mirrors
/
mmselfsup
mirror of https://github.com/open-mmlab/mmselfsup.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Bump version to v0.7.0 (#229) 

* [Enhance] add pre-commit hook for algo-readme and copyright (#213)

* [Enhance] add test windows in workflows (#215)

* [Enhance] add test windows in workflows

* fix lint

* add optional requirements

* add try-except judgement

* add opencv installation in windows test steps

* fix path error on windows

* update

* update path

* update

* add pytest skip for algorithm test

* update requirements/runtime.txt

* update pytest skip

* [Docs] translate 0_config.md into Chinese (#216)

* [Docs] translate 0_config.md into Chinese

* [Fix] fix format description in 0_config.md

* Update: 0_config.md

* [Fix] fix tsne 'no `init_cfg`' error (#222)

* [Fix] fix tsne 'no init_cfg' and pool_type errors

* [Refactor] fix linting of tsne vis

* [Docs] reorganizing OpenMMLab projects and update algorithms in readme (#219)

* [Docs] reorganizing OpenMMLab projects and update algorithms in readme

* using small letters

* fix typo

* [Fix] fix image channel bgr/rgb bug and update benchmarks (#210)

* [Fix] fix image channel bgr/rgb bug

* update model zoo

* update readme and metafile

* [Fix] fix typo

* [Fix] fix typo

* [Fix] fix lint

* modify Places205 directory according to the downloaded dataset

* update results

* [Fix] Fix the bug when using prefetch under multi-view methods, e.g., DenseCL (#218)

* fig bug for prefetch_loader under multi-view setting

* fix lint problem

Co-authored-by: liming <liming.ai@bytedance.com>

* [Feature]: MAE official (#221)

* [Feature]: MAE single image pre-training

* [Fix]: Fix config

* [Fix]: Fix dataset link

* [Feature]: Add run

* [Refactor]: Delete spot

* [Feature]: ignore nohup output file

* [Feature]: Add auto script to generate run cmd

* [Refactor]: Refactor mae config file

* [Feature]: sz20 settings

* [Feature]: Add auto resume

* [Fix]: Fix lint

* [Feature]: Make git ignore txt

* [Refactor]: Delete gpus in script

* [Fix]: Make generate_cmd to add --async

* [Feature]: Initial version of Vit fine-tune

* [Fix]: Add 1424 specific settings

* [Fix]: Fix missing file client bug for 1424

* [Feature]: 1424 customized settings

* [Fix]: Make drop in eval to False

* [Feature]: Change the finetune and pre-training settings

* [Feature]: Add debug setting

* [Refactor]: Refactor the model

* [Feature]: Customized settings

* [Feature]: Add A100 settings

* [Fix]: Change mae to imagenet

* [Feature]: Change mae pretrain num workers to 32

* [Feature]: Change num workers to 16

* [Feature]: Add A100 setting for pre_release ft version

* [Feature]: Add img_norm_cfg

* [Fix]: Fix mae cls test missing logits bug

* [Fix]: Fix mae cls head bias initialize to zero

* [Feature]: Rename mae config name

* [Feature]: Add MAE README.md

* [Fix]: Fix lint

* [Feature]: Fix typo

* [Fix]: Fix typo

* [Feature]: Fix invalid link

* [Fix]: Fix finetune config file name

* [Feature]: Official pretrain v1

* [Feature]: Change log interval to 100

* [Feature]: pretrain 1600 epochs

* [Fix]: Change encoder num head to 12

* [Feature]: Mix precision

* [Feature]: Add default value to random masking

* [Feature]: Official MAE finetune

* [Feature]: Finetune img per gpu 32

* [Feature]: Add multi machine training for lincls

* [Fix]: Fix lincls master port master addr

* [Feature]: Change img per gpu to 128

* [Feature]: Add linear eval and Refactor

* [Fix]: Fix debug mode

* [Fix]: Delete MAE dataset in __init__.py

* [Feature]: normalize pixel for mae

* [Fix]: Fix lint

* [Feature]: LARS for linear eval

* [Feature]: Add lars for mae linear eval

* [Feature]: Change mae linear lars num workers to 32

* [Feature]: Change mae linear lars num workers to 8

* [Feature]: log every 25 iter for mae linear eval lars

* [Feature]: Add 1600 epoch and 800 epoch pretraining

* [Fix]: Change linear eval to 902

* [Fix]: Add random flip to linear eval

* [Fix]: delete fp16 in mae

* [Refactor]: Change backbone to mmcls

* [Fix]: Align finetune settings

* [Fix]: replace timm trunc_normal with mmcv trunc_normal

* [Fix]: Change finetune layer_decay to 0.65

* [Fix]: Delete pretrain last norm when global_pooling

* [Fix]: set requires_grad of norm1 to False

* [Fix]: delete norm1

* [Fix]: Fix docstring bug

* [Fix]: Fix lint

* [Fix]: Add external link

* [Fix]: Delete auto_resume and reformat config readme.

* [Fix]: Fix pytest bug

* [Fix]: Fix lint

* [Refactor]: Rename filename

* [Feature]: Add docstring

* [Fix]: Rename config file name

* [Fix]: Fix name inconsistency bug

* [Fix]: Change the default value of persistent_worker in builder to True

* [Fix]: Change the default value of CPUS_PER_TASK to 5

* [Fix]: Add a blank line to line136 in tools/train.py

* [Fix]: Fix MAE algorithm docstring format and add paper name and url

* [Feature]: Add MAE paper name and link, and store mae teaser on github

* [Refactor]: Delete mae.png

* [Fix]: Fix config file name”

* [Fix]: Fix name bug

* [Refactor]: Change default GPUS to 8

* [Fix]: Abandon change to drop_last

* [Fix]: Fix docstring in mae algorithm

* [Fix]: Fix lint

* [Fix]: Fix lint

* [Fix]: Fix mae finetune algo type bug

* [Feature]: Add unit test for algorithm

* [Feature]: Add unit test for remaining parts

* [Fix]: Fix lint

* [Fix]: Fix typo

* [Fix]: Delete some unnecessary modification in gitignore

* [Feature]: Change finetune setting in mae algo to mixup setting

* [Fix]: Change norm_pix_loss to norm_pix in pretrain head

* [Fix]: Delete modification in dist_train_linear.sh

* [Refactor]: Delete global pool in mae_cls_vit.py

* [Fix]: Change finetune param to mixup in test_mae_classification

* [Fix]: Change norm_pix_loss to norm_pix of mae_pretrain_head in unit test

* [Fix]: Change norm_pix_loss to norm_pix in unit test

* [Refactor]: Create init_weights for mae_finetune_head and mae_linprobe_head

* [Refactor]: Construct 2d sin-cosine position embedding using torch

* [Refactor]: Using classification and using mixup from mmcls

* [Fix]: Fix lint

* [Fix]: Add False to finetune mae linprobe‘
“

* [Fix]: Set drop_last to False

* [Fix]: Fix MAE finetune layerwise lr bug

* [Refactor]: Delete redundant MAE when registering MAE

* [Refactor]: Split initialize_weights in MAE to submodules

* [Fix]: Change the min_lr of mae pretrain to 0.0

* [Refactor]: Delete  unused _init_weights in mae_cls_vit

* [Refactor]: Change MAE cls vit to a more general name

* [Feature]: Add Epoch Fix cosine annealing lr updater

* [Fix]: Fix lint

* [Feature]: Add layer wise lr decay in optimizer constructor

* [Fix]: Fix lint

* [Fix]: Fix set layer wise lr decay bug

* [Fix]: Fix UT for MAE

* [Fix]: Fix lint

* [Fix]: update algorithm readme format for MAE

* [Fix]: Fix isort

* [Fix]: Add Returns inmae_pretrain_vit

* [Fix]: Change bgr to rgb

* [Fix]: Change norm pix to True

* [Fix]: Use cls_token to linear prob

* [Fix]: Delete mixup.py

* [Fix]: Fix MAE readme

* [Feature]: Delete linprobe

* [Refactor]: Merge MAE head into one file

* [Fix]: Fix lint

* [Fix]: rename mae_pretrain_head to mae_head

* [Fix]: Fix import error in __init__.py

* [Feature]: skip MAE algo UT when running on windows

* [Fix]: Fix UT bug

* [Feature]: Update model_zoo

* [Fix]: Rename MAE pretrain model name

* [Fix]: Delete mae ft prefix

* [Feature]: Change b to base

* [Refactor]: Change b in MAE pt config to base

* [Fix]: Fix typo in docstring

* [Fix]: Fix name bug

* [Feature]: Add new constructor for MAE finetune

* [Fix]: Fix model_zoo link

* [Fix]: Skip UT for MAE

* [Fix]: Change fixed channel order to param

Co-authored-by: LIU Yuan <liuyuuan@pjlab.org.cn>
Co-authored-by: liu yuan <liuyuan@pjlab.org.cn>

* [Feature]: Add diff seeds to diff ranks and set torch seed in worker_init_fn (#228)

* [Feature]: Add set diff seeds to diff ranks

* [Fix]: Set diff seed to diff workers

* Bump version to v0.7.0 (#227)

* Bump version to v0.7.0

* [Docs] update readme

Co-authored-by: wang11wang <95845452+wang11wang@users.noreply.github.com>
Co-authored-by: Liangyu Chen <45140242+c-liangyu@users.noreply.github.com>
Co-authored-by: Ming Li <73068772+mitming@users.noreply.github.com>
Co-authored-by: liming <liming.ai@bytedance.com>
Co-authored-by: Yuan Liu <30762564+YuanLiuuuuuu@users.noreply.github.com>
Co-authored-by: LIU Yuan <liuyuuan@pjlab.org.cn>
Co-authored-by: liu yuan <liuyuan@pjlab.org.cn>
pull/235/head v0.7.0
Yixiao Fang 2022-03-04 13:43:49 +08:00 committed by GitHub
parent af331b043f
commit 47f6feb925
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
110 changed files with 2093 additions and 609 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

37
.github/workflows/build.yml vendored
View File

@ -99,3 +99,40 @@ jobs:
env_vars: OS,PYTHON
name: codecov-umbrella
fail_ci_if_error: false
test_windows:
runs-on: ${{ matrix.os }}
strategy:
matrix:
os: [windows-2022]
python: [3.8]
platform: [cpu, cu111]
steps:
- uses: actions/checkout@v2
- name: Set up Python ${{ matrix.python }}
uses: actions/setup-python@v2
with:
python-version: ${{ matrix.python }}
- name: Upgrade pip
run: pip install pip --upgrade --user
- name: Install PyTorch
# As a complement to Linux CI, we test on PyTorch LTS version
run: pip install torch==1.8.2+${{ matrix.platform }} torchvision==0.9.2+${{ matrix.platform }} -f https://download.pytorch.org/whl/lts/1.8/torch_lts.html
- name: Install OpenCV
run: |
pip install opencv-python>=3
- name: Install MMCV
run: |
pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cpu/torch1.8/index.html --only-binary mmcv-full
- name: Install unittest dependencies
run: |
pip install -r requirements.txt
- name: Build and install
run: pip install -e .
- name: Run unittests
run: |
coverage run --branch --source mmselfsup -m pytest tests/
- name: Generate coverage report
run: |
coverage xml
coverage report -m --omit="mmselfsup/apis/*"

6
.pre-commit-config.yaml
View File

@ -40,3 +40,9 @@ repos:
hooks:
- id: docformatter
args: ["--in-place", "--wrap-descriptions", "79"]
- repo: https://github.com/open-mmlab/pre-commit-hooks
rev: v0.2.0
hooks:
- id: check-algo-readme
- id: check-copyright
args: ["mmselfsup", "tools", "tests"]

26
README.md
View File

@ -66,7 +66,13 @@ This project is released under the [Apache 2.0 license](LICENSE).
## ChangeLog
MMSelfSup **v0.6.0** was released in 02/02/2022.
MMSelfSup **v0.7.0** was released in 03/03/2022.
Highlights of the new version:
* Support **MAE**
* Add **Places205** benchmarks
* Add test Windows in workflows
Please refer to [changelog.md](docs/en/changelog.md) for details and release history.
@ -92,6 +98,7 @@ Supported algorithms:
- [x] [DenseCL (CVPR'2021)](https://arxiv.org/abs/2011.09157)
- [x] [SimSiam (CVPR'2021)](https://arxiv.org/abs/2011.10566)
- [x] [MoCo v3 (ICCV'2021)](https://arxiv.org/abs/2104.02057)
- [x] [MAE](https://arxiv.org/abs/2111.06377)
More algorithms are in our plan.
@ -158,20 +165,21 @@ Remarks:
## Projects in OpenMMLab
- [MMCV](https://github.com/open-mmlab/mmcv): OpenMMLab foundational library for computer vision.
- [MIM](https://github.com/open-mmlab/mim): MIM Installs OpenMMLab Packages.
- [MIM](https://github.com/open-mmlab/mim): MIM installs OpenMMLab packages.
- [MMClassification](https://github.com/open-mmlab/mmclassification): OpenMMLab image classification toolbox and benchmark.
- [MMDetection](https://github.com/open-mmlab/mmdetection): OpenMMLab detection toolbox and benchmark.
- [MMDetection3D](https://github.com/open-mmlab/mmdetection3d): OpenMMLab's next-generation platform for general 3D object detection.
- [MMRotate](https://github.com/open-mmlab/mmrotate): OpenMMLab rotated object detection toolbox and benchmark.
- [MMSegmentation](https://github.com/open-mmlab/mmsegmentation): OpenMMLab semantic segmentation toolbox and benchmark.
- [MMAction2](https://github.com/open-mmlab/mmaction2): OpenMMLab's next-generation action understanding toolbox and benchmark.
- [MMTracking](https://github.com/open-mmlab/mmtracking): OpenMMLab video perception toolbox and benchmark.
- [MMOCR](https://github.com/open-mmlab/mmocr): OpenMMLab text detection, recognition, and understanding toolbox.
- [MMPose](https://github.com/open-mmlab/mmpose): OpenMMLab pose estimation toolbox and benchmark.
- [MMEditing](https://github.com/open-mmlab/mmediting): OpenMMLab image and video editing toolbox.
- [MMOCR](https://github.com/open-mmlab/mmocr): OpenMMLab toolbox for text detection, recognition and understanding.
- [MMGeneration](https://github.com/open-mmlab/mmgeneration): OpenMMlab toolkit for generative models.
- [MMFlow](https://github.com/open-mmlab/mmflow): OpenMMLab optical flow toolbox and benchmark.
- [MMFewShot](https://github.com/open-mmlab/mmfewshot): OpenMMLab few shot learning toolbox and benchmark.
- [MMHuman3D](https://github.com/open-mmlab/mmhuman3d): OpenMMLab 3D human parametric model toolbox and benchmark.
- [MMSelfSup](https://github.com/open-mmlab/mmselfsup): OpenMMLab self-supervised learning toolbox and benchmark.
- [MMRazor](https://github.com/open-mmlab/mmrazor): OpenMMLab model compression toolbox and benchmark.
- [MMFewShot](https://github.com/open-mmlab/mmfewshot): OpenMMLab fewshot learning toolbox and benchmark.
- [MMAction2](https://github.com/open-mmlab/mmaction2): OpenMMLab's next-generation action understanding toolbox and benchmark.
- [MMTracking](https://github.com/open-mmlab/mmtracking): OpenMMLab video perception toolbox and benchmark.
- [MMFlow](https://github.com/open-mmlab/mmflow): OpenMMLab optical flow toolbox and benchmark.
- [MMEditing](https://github.com/open-mmlab/mmediting): OpenMMLab image and video editing toolbox.
- [MMGeneration](https://github.com/open-mmlab/mmgeneration): OpenMMLab image and video generative models toolbox.
- [MMDeploy](https://github.com/open-mmlab/mmdeploy): OpenMMLab model deployment framework.

36
README_zh-CN.md
View File

@ -64,7 +64,13 @@ MMSelfSup 是一个基于 PyTorch 实现的开源自监督表征学习工具箱
## 更新日志
最新的 **v0.6.0** 版本已经在 2022.02.02 发布。
最新的 **v0.7.0** 版本已经在 2022.03.03 发布。
新版本亮点：
* 支持 **MAE**
* 增加 **Places205** 下游基准测试
* 增加 Windows 测试
请参考 [更新日志](docs/zh_cn/changelog.md) 获取更多细节和历史版本信息。
@ -91,6 +97,7 @@ MMSelfSup 和 OpenSelfSup 的不同点写在 [对比文档](docs/en/compatibilit
- [x] [DenseCL (CVPR'2021)](https://arxiv.org/abs/2011.09157)
- [x] [SimSiam (CVPR'2021)](https://arxiv.org/abs/2011.10566)
- [x] [MoCo v3 (ICCV'2021)](https://arxiv.org/abs/2104.02057)
- [x] [MAE](https://arxiv.org/abs/2111.06377)
更多的算法实现已经在我们的计划中。
@ -152,22 +159,23 @@ MMSelfSup 是一款由不同学校和公司共同贡献的开源项目，我们
## OpenMMLab 的其他项目
- [MMCV](https://github.com/open-mmlab/mmcv): OpenMMLab 计算机视觉基础库
- [MIM](https://github.com/open-mmlab/mim): MIM 是 OpenMMLab 项目、算法、模型的统一入口
- [MMClassification](https://github.com/open-mmlab/mmclassification): OpenMMLab 图像分类工具箱与测试基准
- [MMDetection](https://github.com/open-mmlab/mmdetection): OpenMMLab 检测工具箱与测试基准
- [MMDetection3D](https://github.com/open-mmlab/mmdetection3d): OpenMMLab 新一代通用3D目标检测平台
- [MMSegmentation](https://github.com/open-mmlab/mmsegmentation): OpenMMLab 语义分割工具箱与测试基准
- [MMAction2](https://github.com/open-mmlab/mmaction2): OpenMMLab 新一代视频理解工具箱与测试基准
- [MMTracking](https://github.com/open-mmlab/mmtracking): OpenMMLab 一体化视频目标感知平台
- [MMPose](https://github.com/open-mmlab/mmpose): OpenMMLab 姿态估计工具箱与测试基准
- [MMEditing](https://github.com/open-mmlab/mmediting): OpenMMLab 图像视频编辑工具箱
- [MMOCR](https://github.com/open-mmlab/mmocr): OpenMMLab 全流程文字检测识别理解工具包
- [MMGeneration](https://github.com/open-mmlab/mmgeneration): OpenMMLab 新一代生成模型工具箱
- [MMFlow](https://github.com/open-mmlab/mmflow): OpenMMLab 光流估计工具箱与测试基准
- [MMFewShot](https://github.com/open-mmlab/mmfewshot): OpenMMLab 少样本学习工具箱与测试基准
- [MIM](https://github.com/open-mmlab/mim): MIM 是 OpenMMlab 项目、算法、模型的统一入口
- [MMClassification](https://github.com/open-mmlab/mmclassification): OpenMMLab 图像分类工具箱
- [MMDetection](https://github.com/open-mmlab/mmdetection): OpenMMLab 目标检测工具箱
- [MMDetection3D](https://github.com/open-mmlab/mmdetection3d): OpenMMLab 新一代通用 3D 目标检测平台
- [MMRotate](https://github.com/open-mmlab/mmrotate): OpenMMLab 旋转框检测工具箱与测试基准
- [MMSegmentation](https://github.com/open-mmlab/mmsegmentation): OpenMMLab 语义分割工具箱
- [MMOCR](https://github.com/open-mmlab/mmocr): OpenMMLab 全流程文字检测识别理解工具箱
- [MMPose](https://github.com/open-mmlab/mmpose): OpenMMLab 姿态估计工具箱
- [MMHuman3D](https://github.com/open-mmlab/mmhuman3d): OpenMMLab 人体参数化模型工具箱与测试基准
- [MMSelfSup](https://github.com/open-mmlab/mmselfsup): OpenMMLab 自监督学习工具箱与测试基准
- [MMRazor](https://github.com/open-mmlab/mmrazor): OpenMMLab 模型压缩工具箱与测试基准
- [MMFewShot](https://github.com/open-mmlab/mmfewshot): OpenMMLab 少样本学习工具箱与测试基准
- [MMAction2](https://github.com/open-mmlab/mmaction2): OpenMMLab 新一代视频理解工具箱
- [MMTracking](https://github.com/open-mmlab/mmtracking): OpenMMLab 一体化视频目标感知平台
- [MMFlow](https://github.com/open-mmlab/mmflow): OpenMMLab 光流估计工具箱与测试基准
- [MMEditing](https://github.com/open-mmlab/mmediting): OpenMMLab 图像视频编辑工具箱
- [MMGeneration](https://github.com/open-mmlab/mmgeneration): OpenMMLab 图片视频生成模型工具箱
- [MMDeploy](https://github.com/open-mmlab/mmdeploy): OpenMMLab 模型部署框架
## 欢迎加入 OpenMMLab 社区

12
configs/benchmarks/classification/_base_/datasets/places205.py
View File

@ -31,8 +31,10 @@ data = dict(
type=dataset_type,
data_source=dict(
type=data_source,
data_prefix='data/Places205/train',
ann_file='data/Places205/meta/train.txt',
data_prefix= # noqa: E251
'data/Places205/data/vision/torralba/deeplearning/images256/',
ann_file= # noqa: E251
'data/Places205/trainvalsplit_places205/train_places205.csv',
),
pipeline=train_pipeline,
prefetch=prefetch),
@ -40,8 +42,10 @@ data = dict(
type=dataset_type,
data_source=dict(
type=data_source,
data_prefix='data/Places205/val',
ann_file='data/Places205/meta/val.txt',
data_prefix= # noqa: E251
'data/Places205/data/vision/torralba/deeplearning/images256/',
ann_file= # noqa: E251
'data/Places205/trainvalsplit_places205/val_places205.csv',
),
pipeline=test_pipeline,
prefetch=prefetch))

17
configs/benchmarks/classification/_base_/models/vit-base-p16_ft.py 100644
View File

@ -0,0 +1,17 @@
model = dict(
type='Classification',
backbone=dict(
type='MIMVisionTransformer',
arch='b',
patch_size=16,
drop_path_rate=0.1,
final_norm=False),
head=dict(
type='MAEFinetuneHead',
num_classes=1000,
embed_dim=768,
label_smooth_val=0.1),
train_cfg=dict(augments=[
dict(type='BatchMixup', alpha=0.8, num_classes=1000, prob=0.5),
dict(type='BatchCutMix', alpha=1.0, num_classes=1000, prob=0.5)
]))

9
configs/benchmarks/classification/_base_/models/vit-base-p16_linprobe.py 100644
View File

@ -0,0 +1,9 @@
model = dict(
type='Classification',
backbone=dict(
type='MIMVisionTransformer',
arch='b',
patch_size=16,
final_norm=True,
finetune=False),
head=dict(type='MAELinprobeHead', num_classes=1000, embed_dim=768))

14
configs/benchmarks/classification/_base_/schedules/adamw_coslr-100e_in1k.py 100644
View File

@ -0,0 +1,14 @@
# optimizer
optimizer = dict(type='AdamW', lr=1e-3, betas=(0.9, 0.999), weight_decay=0.05)
# learning policy
lr_config = dict(
policy='CosineAnnealing',
min_lr=0.,
warmup='linear',
warmup_iters=5,
warmup_ratio=1e-4, # cannot be 0
warmup_by_epoch=True)
# runtime settings
runner = dict(type='EpochBasedRunner', max_epochs=100)

67
configs/benchmarks/classification/imagenet/vit-b-p16_ft-8xb128-coslr-100e_in1k.py 100644
View File

@ -0,0 +1,67 @@
_base_ = [
'../_base_/models/vit-base-p16_ft.py',
'../_base_/datasets/imagenet.py',
'../_base_/schedules/adamw_coslr-100e_in1k.py',
'../_base_/default_runtime.py',
]
# dataset
img_norm_cfg = dict(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
train_pipeline = [
dict(
type='RandomAug',
input_size=224,
color_jitter=None,
auto_augment='rand-m9-mstd0.5-inc1',
interpolation='bicubic',
re_prob=0.25,
re_mode='pixel',
re_count=1,
mean=(0.485, 0.456, 0.406),
std=(0.229, 0.224, 0.225))
]
test_pipeline = [
dict(type='Resize', size=256, interpolation=3),
dict(type='CenterCrop', size=224),
dict(type='ToTensor'),
dict(type='Normalize', **img_norm_cfg)
]
data = dict(
samples_per_gpu=128,
drop_last=False,
workers_per_gpu=32,
train=dict(pipeline=train_pipeline),
val=dict(pipeline=test_pipeline))
# model
model = dict(backbone=dict(init_cfg=dict()))
# optimizer
optimizer = dict(
lr=1e-3 * 1024 / 256,
paramwise_options={
'norm': dict(weight_decay=0.),
'bias': dict(weight_decay=0.),
'pos_embed': dict(weight_decay=0.),
'cls_token': dict(weight_decay=0.)
},
constructor='MAEFtOptimizerConstructor',
layer_decay=0.65)
# learning policy
lr_config = dict(
policy='StepFixCosineAnnealing',
min_lr=1e-6,
warmup='linear',
warmup_iters=5,
warmup_ratio=1e-4,
warmup_by_epoch=True,
by_epoch=False)
# runtime
checkpoint_config = dict(interval=1, max_keep_ckpts=3, out_dir='')
persistent_workers = True
log_config = dict(
interval=100, hooks=[
dict(type='TextLoggerHook'),
])

30
configs/selfsup/_base_/datasets/imagenet_mae.py 100644
View File

@ -0,0 +1,30 @@
# dataset settings
data_source = 'ImageNet'
dataset_type = 'SingleViewDataset'
img_norm_cfg = dict(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
train_pipeline = [
dict(
type='RandomResizedCrop', size=224, scale=(0.2, 1.0), interpolation=3),
dict(type='RandomHorizontalFlip')
]
# prefetch
prefetch = False
if not prefetch:
train_pipeline.extend(
[dict(type='ToTensor'),
dict(type='Normalize', **img_norm_cfg)])
# dataset summary
data = dict(
imgs_per_gpu=128,
workers_per_gpu=8,
train=dict(
type=dataset_type,
data_source=dict(
type=data_source,
data_prefix='data/imagenet/train',
ann_file='data/imagenet/meta/train.txt',
),
pipeline=train_pipeline,
prefetch=prefetch))

15
configs/selfsup/_base_/models/mae_vit-base-p16.py 100644
View File

@ -0,0 +1,15 @@
# model settings
model = dict(
type='MAE',
backbone=dict(type='MAEViT', arch='b', patch_size=16, mask_ratio=0.75),
neck=dict(
type='MAEPretrainDecoder',
patch_size=16,
in_chans=3,
embed_dim=768,
decoder_embed_dim=512,
decoder_depth=8,
decoder_num_heads=16,
mlp_ratio=4.,
),
head=dict(type='MAEPretrainHead', norm_pix=True, patch_size=16))

15
configs/selfsup/_base_/schedules/adamw_coslr-200e_in1k.py 100644
View File

@ -0,0 +1,15 @@
# optimizer
optimizer = dict(type='AdamW', lr=1.5e-4, betas=(0.9, 0.95), weight_decay=0.05)
optimizer_config = dict() # grad_clip, coalesce, bucket_size_mb
# learning policy
lr_config = dict(
policy='CosineAnnealing',
min_lr=0.,
warmup='linear',
warmup_iters=40,
warmup_ratio=1e-4, # cannot be 0
warmup_by_epoch=True)
# runtime settings
runner = dict(type='EpochBasedRunner', max_epochs=300)

51
configs/selfsup/byol/README.md
View File

@ -14,7 +14,7 @@
## Results and Models
**Back to [model_zoo.md](../../../docs/en/model_zoo.md) to download models.**
**Back to [model_zoo.md](https://github.com/open-mmlab/mmselfsup/blob/master/docs/en/model_zoo.md) to download models.**
In this page, we provide benchmarks as much as possible to evaluate our pre-trained models. If not mentioned, all models were trained on ImageNet1k dataset.
@ -28,19 +28,28 @@ The **Best Layer** indicates that the best results are obtained from which layer
Besides, k=1 to 96 indicates the hyper-parameter of Low-shot SVM.
| Self-Supervised Config | Best Layer | SVM | k=1 | k=2 | k=4 | k=8 | k=16 | k=32 | k=64 | k=96 |
| ----------------------------------------------------------------------------------- | ---------- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- |
| [resnet50_8xb32-accum16-coslr-200e](byol_resnet50_8xb32-accum16-coslr-200e_in1k.py) | feature5 | 86.31 | 45.37 | 56.83 | 68.47 | 74.12 | 78.30 | 81.53 | 83.56 | 84.73 |
| Self-Supervised Config | Best Layer | SVM | k=1 | k=2 | k=4 | k=8 | k=16 | k=32 | k=64 | k=96 |
| ------------------------------------------------------------------------------------------------------------------------------------------------------------ | ---------- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- |
| [resnet50_8xb32-accum16-coslr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/byol/byol_resnet50_8xb32-accum16-coslr-200e_in1k.py) | feature5 | 86.31 | 45.37 | 56.83 | 68.47 | 74.12 | 78.30 | 81.53 | 83.56 | 84.73 |
#### ImageNet Linear Evaluation
The **Feature1 - Feature5** don't have the GlobalAveragePooling, the feature map is pooled to the specific dimensions and then follows a Linear layer to do the classification. Please refer to [resnet50_mhead_8xb32-steplr-90e.py](../../benchmarks/classification/imagenet/resnet50_mhead_8xb32-steplr-90e_in1k.py) for details of config.
The **Feature1 - Feature5** don't have the GlobalAveragePooling, the feature map is pooled to the specific dimensions and then follows a Linear layer to do the classification. Please refer to [resnet50_mhead_8xb32-steplr-90e.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/classification/imagenet/resnet50_mhead_8xb32-steplr-90e_in1k.py) for details of config.
The **AvgPool** result is obtained from Linear Evaluation with GlobalAveragePooling. Please refer to [resnet50_8xb32-steplr-100e_in1k](../../benchmarks/classification/imagenet/resnet50_8xb32-steplr-100e_in1k.py) for details of config.
The **AvgPool** result is obtained from Linear Evaluation with GlobalAveragePooling. Please refer to [resnet50_8xb32-steplr-100e_in1k.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/classification/imagenet/resnet50_8xb32-steplr-100e_in1k.py) for details of config.
| Self-Supervised Config | Feature1 | Feature2 | Feature3 | Feature4 | Feature5 | AvgPool |
| ----------------------------------------------------------------------------------- | -------- | -------- | -------- | -------- | -------- | ------- |
| [resnet50_8xb32-accum16-coslr-200e](byol_resnet50_8xb32-accum16-coslr-200e_in1k.py) | 15.59 | 35.16 | 47.37 | 62.86 | 71.62 | 67.68 |
| Self-Supervised Config | Feature1 | Feature2 | Feature3 | Feature4 | Feature5 | AvgPool |
| ------------------------------------------------------------------------------------------------------------------------------------------------------------ | -------- | -------- | -------- | -------- | -------- | ------- |
| [resnet50_8xb32-accum16-coslr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/byol/byol_resnet50_8xb32-accum16-coslr-200e_in1k.py) | 15.59 | 35.16 | 47.37 | 62.86 | 71.62 | 67.55 |
#### Places205 Linear Evaluation
The **Feature1 - Feature5** don't have the GlobalAveragePooling, the feature map is pooled to the specific dimensions and then follows a Linear layer to do the classification. Please refer to [resnet50_mhead_8xb32-steplr-28e_places205.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/classification/places205/resnet50_mhead_8xb32-steplr-28e_places205.py) for details of config.
| Self-Supervised Config | Feature1 | Feature2 | Feature3 | Feature4 | Feature5 |
| ------------------------------------------------------------------------------------------------------------------------------------------------------------ | -------- | -------- | -------- | -------- | -------- |
| [resnet50_8xb32-accum16-coslr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/byol/byol_resnet50_8xb32-accum16-coslr-200e_in1k.py) | 21.25 | 36.55 | 43.66 | 50.74 | 53.82 |
| [resnet50_8xb32-accum16-coslr-300e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/byol/byol_resnet50_8xb32-accum16-coslr-300e_in1k.py) | 21.18 | 36.68 | 43.42 | 51.04 | 54.06 |
### Detection
@ -48,19 +57,19 @@ The detection benchmarks includes 2 downstream task datasets, **Pascal VOC 2007
#### Pascal VOC 2007 + 2012
Please refer to [faster_rcnn_r50_c4_mstrain_24k_voc0712.py](../../benchmarks/mmdetection/voc0712/faster_rcnn_r50_c4_mstrain_24k_voc0712.py) for details of config.
Please refer to [faster_rcnn_r50_c4_mstrain_24k_voc0712.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/mmdetection/voc0712/faster_rcnn_r50_c4_mstrain_24k_voc0712.py) for details of config.
| Self-Supervised Config | AP50 |
| ----------------------------------------------------------------------------------- | ----- |
| [resnet50_8xb32-accum16-coslr-200e](byol_resnet50_8xb32-accum16-coslr-200e_in1k.py) | 80.35 |
| Self-Supervised Config | AP50 |
| ------------------------------------------------------------------------------------------------------------------------------------------------------------ | ----- |
| [resnet50_8xb32-accum16-coslr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/byol/byol_resnet50_8xb32-accum16-coslr-200e_in1k.py) | 80.35 |
#### COCO2017
Please refer to [mask_rcnn_r50_fpn_mstrain_1x_coco.py](../../benchmarks/mmdetection/coco/mask_rcnn_r50_fpn_mstrain_1x_coco.py) for details of config.
Please refer to [mask_rcnn_r50_fpn_mstrain_1x_coco.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/mmdetection/coco/mask_rcnn_r50_fpn_mstrain_1x_coco.py) for details of config.
| Self-Supervised Config | mAP(Box) | AP50(Box) | AP75(Box) | mAP(Mask) | AP50(Mask) | AP75(Mask) |
| ----------------------------------------------------------------------------------- | -------- | --------- | --------- | --------- | ---------- | ---------- |
| [resnet50_8xb32-accum16-coslr-200e](byol_resnet50_8xb32-accum16-coslr-200e_in1k.py) | 40.9 | 61.0 | 44.6 | 36.8 | 58.1 | 39.5 |
| Self-Supervised Config | mAP(Box) | AP50(Box) | AP75(Box) | mAP(Mask) | AP50(Mask) | AP75(Mask) |
| ------------------------------------------------------------------------------------------------------------------------------------------------------------ | -------- | --------- | --------- | --------- | ---------- | ---------- |
| [resnet50_8xb32-accum16-coslr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/byol/byol_resnet50_8xb32-accum16-coslr-200e_in1k.py) | 40.9 | 61.0 | 44.6 | 36.8 | 58.1 | 39.5 |
### Segmentation
@ -68,11 +77,11 @@ The segmentation benchmarks includes 2 downstream task datasets, **Cityscapes**
#### Pascal VOC 2012 + Aug
Please refer to [fcn_r50-d8_512x512_20k_voc12aug.py](../../benchmarks/mmsegmentation/voc12aug/fcn_r50-d8_512x512_20k_voc12aug.py) for details of config.
Please refer to [fcn_r50-d8_512x512_20k_voc12aug.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/mmsegmentation/voc12aug/fcn_r50-d8_512x512_20k_voc12aug.py) for details of config.
| Self-Supervised Config | mIOU |
| ----------------------------------------------------------------------------------- | ----- |
| [resnet50_8xb32-accum16-coslr-200e](byol_resnet50_8xb32-accum16-coslr-200e_in1k.py) | 67.16 |
| Self-Supervised Config | mIOU |
| ------------------------------------------------------------------------------------------------------------------------------------------------------------ | ----- |
| [resnet50_8xb32-accum16-coslr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/byol/byol_resnet50_8xb32-accum16-coslr-200e_in1k.py) | 67.16 |
## Citation

16
configs/selfsup/byol/metafile.yml
View File

@ -23,6 +23,18 @@ Models:
- Task: Self-Supervised Image Classification
Dataset: ImageNet-1k
Metrics:
Top 1 Accuracy: 67.68
Top 1 Accuracy: 67.55
Config: configs/selfsup/byol/byol_resnet50_8xb32-accum16-coslr-200e_in1k.py
Weights: https://download.openmmlab.com/mmselfsup/byol/byol_resnet50_8xb32-accum16-coslr-200e_in1k_20211213-30dbaef1.pth
Weights: https://download.openmmlab.com/mmselfsup/byol/byol_resnet50_8xb32-accum16-coslr-200e_in1k_20220225-5c8b2c2e.pth
- Name: byol_resnet50_8xb32-accum16-coslr-300e_in1k
In Collection: BYOL
Metadata:
Epochs: 300
Batch Size: 256
Results:
- Task: Self-Supervised Image Classification
Dataset: ImageNet-1k
Metrics:
Top 1 Accuracy: 68.55
Config: configs/selfsup/byol/byol_resnet50_8xb32-accum16-coslr-200e_in1k.py
Weights: https://download.openmmlab.com/mmselfsup/byol/byol_resnet50_8xb32-accum16-coslr-300e_in1k_20220225-a0daa54a.pth

26
configs/selfsup/deepcluster/README.md
View File

@ -14,7 +14,7 @@ Clustering is a class of unsupervised learning methods that has been extensively
## Results and Models
**Back to [model_zoo.md](../../../docs/en/model_zoo.md) to download models.**
**Back to [model_zoo.md](https://github.com/open-mmlab/mmselfsup/blob/master/docs/en/model_zoo.md) to download models.**
In this page, we provide benchmarks as much as possible to evaluate our pre-trained models. If not mentioned, all models were trained on ImageNet1k dataset.
@ -28,19 +28,27 @@ The **Best Layer** indicates that the best results are obtained from which layer
Besides, k=1 to 96 indicates the hyper-parameter of Low-shot SVM.
| Self-Supervised Config | Best Layer | SVM | k=1 | k=2 | k=4 | k=8 | k=16 | k=32 | k=64 | k=96 |
| ---------------------------------------------------------------------------------------- | ---------- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- |
| [sobel_resnet50_8xb64-steplr-200e](deepcluster-sobel_resnet50_8xb64-steplr-200e_in1k.py) | feature5 | 74.26 | 29.37 | 37.99 | 45.85 | 55.57 | 62.48 | 66.15 | 70.00 | 71.37 |
| Self-Supervised Config | Best Layer | SVM | k=1 | k=2 | k=4 | k=8 | k=16 | k=32 | k=64 | k=96 |
| ------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | ---------- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- |
| [sobel_resnet50_8xb64-steplr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/deepcluster/deepcluster-sobel_resnet50_8xb64-steplr-200e_in1k.py) | feature5 | 74.26 | 29.37 | 37.99 | 45.85 | 55.57 | 62.48 | 66.15 | 70.00 | 71.37 |
#### ImageNet Linear Evaluation
The **Feature1 - Feature5** don't have the GlobalAveragePooling, the feature map is pooled to the specific dimensions and then follows a Linear layer to do the classification. Please refer to [resnet50_mhead_8xb32-steplr-90e.py](../../benchmarks/classification/imagenet/resnet50_mhead_8xb32-steplr-90e_in1k.py) for details of config.
The **Feature1 - Feature5** don't have the GlobalAveragePooling, the feature map is pooled to the specific dimensions and then follows a Linear layer to do the classification. Please refer to [resnet50_mhead_8xb32-steplr-90e.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/classification/imagenet/resnet50_mhead_8xb32-steplr-90e_in1k.py) for details of config.
The **AvgPool** result is obtained from Linear Evaluation with GlobalAveragePooling. Please refer to [file name]() for details of config.
The **AvgPool** result is obtained from Linear Evaluation with GlobalAveragePooling. Please refer to [resnet50_8xb32-steplr-100e_in1k.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/classification/imagenet/resnet50_8xb32-steplr-100e_in1k.py) for details of config.
| Self-Supervised Config | Feature1 | Feature2 | Feature3 | Feature4 | Feature5 | AvgPool |
| ---------------------------------------------------------------------------------------- | -------- | -------- | -------- | -------- | -------- | ------- |
| [sobel_resnet50_8xb64-steplr-200e](deepcluster-sobel_resnet50_8xb64-steplr-200e_in1k.py) | 12.78 | 30.81 | 43.88 | 57.71 | 51.68 | 46.92 |
| Self-Supervised Config | Feature1 | Feature2 | Feature3 | Feature4 | Feature5 | AvgPool |
| ------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | -------- | -------- | -------- | -------- | -------- | ------- |
| [sobel_resnet50_8xb64-steplr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/deepcluster/deepcluster-sobel_resnet50_8xb64-steplr-200e_in1k.py) | 12.78 | 30.81 | 43.88 | 57.71 | 51.68 | 46.92 |
#### Places205 Linear Evaluation
The **Feature1 - Feature5** don't have the GlobalAveragePooling, the feature map is pooled to the specific dimensions and then follows a Linear layer to do the classification. Please refer to [resnet50_mhead_8xb32-steplr-28e_places205.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/classification/places205/resnet50_mhead_8xb32-steplr-28e_places205.py) for details of config.
| Self-Supervised Config | Feature1 | Feature2 | Feature3 | Feature4 | Feature5 |
| ------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | -------- | -------- | -------- | -------- | -------- |
| [sobel_resnet50_8xb64-steplr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/deepcluster/deepcluster-sobel_resnet50_8xb64-steplr-200e_in1k.py) | 18.80 | 33.93 | 41.44 | 47.22 | 42.61 |
## Citation

50
configs/selfsup/densecl/README.md
View File

@ -14,7 +14,7 @@ To date, most existing self-supervised learning methods are designed and optimiz
## Results and Models
**Back to [model_zoo.md](../../../docs/en/model_zoo.md) to download models.**
**Back to [model_zoo.md](https://github.com/open-mmlab/mmselfsup/blob/master/docs/en/model_zoo.md) to download models.**
In this page, we provide benchmarks as much as possible to evaluate our pre-trained models. If not mentioned, all models were trained on ImageNet1k dataset.
@ -28,19 +28,27 @@ The **Best Layer** indicates that the best results are obtained from which layer
Besides, k=1 to 96 indicates the hyper-parameter of Low-shot SVM.
| Self-Supervised Config | Best Layer | SVM | k=1 | k=2 | k=4 | k=8 | k=16 | k=32 | k=64 | k=96 |
| ---------------------------------------------------------------------- | ---------- | ---- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- |
| [resnet50_8xb32-coslr-200e](densecl_resnet50_8xb32-coslr-200e_in1k.py) | feature5 | 82.5 | 42.68 | 50.64 | 61.74 | 68.17 | 72.99 | 76.07 | 79.19 | 80.55 |
| Self-Supervised Config | Best Layer | SVM | k=1 | k=2 | k=4 | k=8 | k=16 | k=32 | k=64 | k=96 |
| -------------------------------------------------------------------------------------------------------------------------------------------------- | ---------- | ---- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- |
| [resnet50_8xb32-coslr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/densecl/densecl_resnet50_8xb32-coslr-200e_in1k.py) | feature5 | 82.5 | 42.68 | 50.64 | 61.74 | 68.17 | 72.99 | 76.07 | 79.19 | 80.55 |
#### ImageNet Linear Evaluation
The **Feature1 - Feature5** don't have the GlobalAveragePooling, the feature map is pooled to the specific dimensions and then follows a Linear layer to do the classification. Please refer to [resnet50_mhead_8xb32-steplr-90e.py](../../benchmarks/classification/imagenet/resnet50_mhead_8xb32-steplr-90e_in1k.py) for details of config.
The **Feature1 - Feature5** don't have the GlobalAveragePooling, the feature map is pooled to the specific dimensions and then follows a Linear layer to do the classification. Please refer to [resnet50_mhead_8xb32-steplr-90e.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/classification/imagenet/resnet50_mhead_8xb32-steplr-90e_in1k.py) for details of config.
The **AvgPool** result is obtained from Linear Evaluation with GlobalAveragePooling. Please refer to [resnet50_8xb32-steplr-100e_in1k](../../benchmarks/classification/imagenet/resnet50_8xb32-steplr-100e_in1k.py) for details of config.
The **AvgPool** result is obtained from Linear Evaluation with GlobalAveragePooling. Please refer to [resnet50_8xb32-steplr-100e_in1k.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/classification/imagenet/resnet50_8xb32-steplr-100e_in1k.py) for details of config.
| Self-Supervised Config | Feature1 | Feature2 | Feature3 | Feature4 | Feature5 | AvgPool |
| ---------------------------------------------------------------------- | -------- | -------- | -------- | -------- | -------- | ------- |
| [resnet50_8xb32-coslr-200e](densecl_resnet50_8xb32-coslr-200e_in1k.py) | 15.86 | 35.47 | 49.46 | 64.06 | 62.95 | 63.34 |
| Self-Supervised Config | Feature1 | Feature2 | Feature3 | Feature4 | Feature5 | AvgPool |
| -------------------------------------------------------------------------------------------------------------------------------------------------- | -------- | -------- | -------- | -------- | -------- | ------- |
| [resnet50_8xb32-coslr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/densecl/densecl_resnet50_8xb32-coslr-200e_in1k.py) | 15.86 | 35.47 | 49.46 | 64.06 | 62.95 | 63.34 |
#### Places205 Linear Evaluation
The **Feature1 - Feature5** don't have the GlobalAveragePooling, the feature map is pooled to the specific dimensions and then follows a Linear layer to do the classification. Please refer to [resnet50_mhead_8xb32-steplr-28e_places205.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/classification/places205/resnet50_mhead_8xb32-steplr-28e_places205.py) for details of config.
| Self-Supervised Config | Feature1 | Feature2 | Feature3 | Feature4 | Feature5 |
| -------------------------------------------------------------------------------------------------------------------------------------------------- | -------- | -------- | -------- | -------- | -------- |
| [resnet50_8xb32-coslr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/densecl/densecl_resnet50_8xb32-coslr-200e_in1k.py) | 21.32 | 36.20 | 43.97 | 51.04 | 50.45 |
### Detection
@ -48,19 +56,19 @@ The detection benchmarks includes 2 downstream task datasets, **Pascal VOC 2007
#### Pascal VOC 2007 + 2012
Please refer to [faster_rcnn_r50_c4_mstrain_24k_voc0712.py](../../benchmarks/mmdetection/voc0712/faster_rcnn_r50_c4_mstrain_24k_voc0712.py) for details of config.
Please refer to [faster_rcnn_r50_c4_mstrain_24k_voc0712.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/mmdetection/voc0712/faster_rcnn_r50_c4_mstrain_24k_voc0712.py) for details of config.
| Self-Supervised Config | AP50 |
| ---------------------------------------------------------------------- | ----- |
| [resnet50_8xb32-coslr-200e](densecl_resnet50_8xb32-coslr-200e_in1k.py) | 82.14 |
| Self-Supervised Config | AP50 |
| -------------------------------------------------------------------------------------------------------------------------------------------------- | ----- |
| [resnet50_8xb32-coslr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/densecl/densecl_resnet50_8xb32-coslr-200e_in1k.py) | 82.14 |
#### COCO2017
Please refer to [mask_rcnn_r50_fpn_mstrain_1x_coco.py](../../benchmarks/mmdetection/coco/mask_rcnn_r50_fpn_mstrain_1x_coco.py) for details of config.
Please refer to [mask_rcnn_r50_fpn_mstrain_1x_coco.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/mmdetection/coco/mask_rcnn_r50_fpn_mstrain_1x_coco.py) for details of config.
| Self-Supervised Config | mAP(Box) | AP50(Box) | AP75(Box) | mAP(Mask) | AP50(Mask) | AP75(Mask) |
| ---------------------------------------------------------------------- | -------- | --------- | --------- | --------- | ---------- | ---------- |
| [resnet50_8xb32-coslr-200e](densecl_resnet50_8xb32-coslr-200e_in1k.py) | | | | | | |
| Self-Supervised Config | mAP(Box) | AP50(Box) | AP75(Box) | mAP(Mask) | AP50(Mask) | AP75(Mask) |
| -------------------------------------------------------------------------------------------------------------------------------------------------- | -------- | --------- | --------- | --------- | ---------- | ---------- |
| [resnet50_8xb32-coslr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/densecl/densecl_resnet50_8xb32-coslr-200e_in1k.py) | | | | | | |
### Segmentation
@ -68,11 +76,11 @@ The segmentation benchmarks includes 2 downstream task datasets, **Cityscapes**
#### Pascal VOC 2012 + Aug
Please refer to [fcn_r50-d8_512x512_20k_voc12aug.py](../../benchmarks/mmsegmentation/voc12aug/fcn_r50-d8_512x512_20k_voc12aug.py) for details of config.
Please refer to [fcn_r50-d8_512x512_20k_voc12aug.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/mmsegmentation/voc12aug/fcn_r50-d8_512x512_20k_voc12aug.py) for details of config.
| Self-Supervised Config | mIOU |
| ---------------------------------------------------------------------- | ----- |
| [resnet50_8xb32-coslr-200e](densecl_resnet50_8xb32-coslr-200e_in1k.py) | 69.47 |
| Self-Supervised Config | mIOU |
| -------------------------------------------------------------------------------------------------------------------------------------------------- | ----- |
| [resnet50_8xb32-coslr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/densecl/densecl_resnet50_8xb32-coslr-200e_in1k.py) | 69.47 |
## Citation

4
configs/selfsup/densecl/metafile.yml
View File

@ -23,6 +23,6 @@ Models:
- Task: Self-Supervised Image Classification
Dataset: ImageNet-1k
Metrics:
Top 1 Accuracy: 63.34
Top 1 Accuracy: 63.62
Config: configs/selfsup/densecl/densecl_resnet50_8xb32-coslr-200e_in1k.py
Weights: https://download.openmmlab.com/mmselfsup/densecl/densecl_resnet50_8xb32-coslr-200e_in1k_20211214-1efb342c.pth
Weights: https://download.openmmlab.com/mmselfsup/densecl/densecl_resnet50_8xb32-coslr-200e_in1k_20220225-8c7808fe.pth

54
configs/selfsup/mae/README.md 100644
View File

@ -0,0 +1,54 @@
# MAE
> [Masked Autoencoders Are Scalable Vision Learners](https://arxiv.org/abs/2111.06377)
<!-- [ALGORITHM] -->
## Abstract
This paper shows that masked autoencoders (MAE) are
scalable self-supervised learners for computer vision. Our
MAE approach is simple: we mask random patches of the
input image and reconstruct the missing pixels. It is based
on two core designs. First, we develop an asymmetric
encoder-decoder architecture, with an encoder that operates only on the
visible subset of patches (without mask tokens), along with a lightweight
decoder that reconstructs the original image from the latent representation
and mask tokens. Second, we find that masking a high proportion
of the input image, e.g., 75%, yields a nontrivial and
meaningful self-supervisory task. Coupling these two designs enables us to
train large models efficiently and effectively: we accelerate
training (by 3× or more) and improve accuracy. Our scalable approach allows
for learning high-capacity models that generalize well: e.g., a vanilla
ViT-Huge model achieves the best accuracy (87.8%) among
methods that use only ImageNet-1K data. Transfer performance in downstream tasks outperforms supervised pretraining and shows promising scaling behavior.
<div align="center">
<img src="https://user-images.githubusercontent.com/30762564/150733959-2959852a-c7bd-4d3f-911f-3e8d8839fe67.png" width="40%"/>
</div>
## Models and Benchmarks
Here, we report the results of the model, which is pre-trained on ImageNet1K
for 400 epochs, the details are below:
| Backbone | Pre-train epoch | Fine-tuning Top-1 | Pre-train Config | Fine-tuning Config | Download |
| :------: | :-------------: | :---------------: | :-------------------------------------------------: | :---------------------------------------------------------------------------------------: | :-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: |
| ViT-B/16 | 400 | 83.1 | [config](./mae_vit-b-p16_8xb512-coslr-400e_in1k.py) | [config](../../benchmarks/classification/imagenet/vit-b-p16_ft-8xb128-coslr-100e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/mae/mae_vit-base-p16_8xb512-coslr-400e_in1k-224_20220223-85be947b.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/mae/mae_vit-base-p16_8xb512-coslr-300e_in1k-224_20220210_140925.log.json) |
## Citation
```bibtex
@article{He2021MaskedAA,
title={Masked Autoencoders Are Scalable Vision Learners},
author={Kaiming He and Xinlei Chen and Saining Xie and Yanghao Li and
Piotr Doll'ar and Ross B. Girshick},
journal={ArXiv},
year={2021},
volume={abs/2111.06377}
}
```

4
configs/selfsup/mae/mae_vit-base-p16_8xb512-coslr-1600e_in1k.py 100644
View File

@ -0,0 +1,4 @@
_base_ = 'mae_vit-base-16_8xb512-coslr-400e_in1k.py'
# schedule
runner = dict(max_epochs=1600)

42
configs/selfsup/mae/mae_vit-base-p16_8xb512-coslr-400e_in1k.py 100644
View File

@ -0,0 +1,42 @@
_base_ = [
'../_base_/models/mae_vit-base-p16.py',
'../_base_/datasets/imagenet_mae.py',
'../_base_/schedules/adamw_coslr-200e_in1k.py',
'../_base_/default_runtime.py',
]
# dataset
data = dict(samples_per_gpu=512, workers_per_gpu=32)
# optimizer
optimizer = dict(
lr=1.5e-4 * 4096 / 256,
paramwise_options={
'norm': dict(weight_decay=0.),
'bias': dict(weight_decay=0.),
'pos_embed': dict(weight_decay=0.),
'mask_token': dict(weight_decay=0.),
'cls_token': dict(weight_decay=0.)
})
optimizer_config = dict()
# learning policy
lr_config = dict(
policy='StepFixCosineAnnealing',
min_lr=0.0,
warmup='linear',
warmup_iters=40,
warmup_ratio=1e-4,
warmup_by_epoch=True,
by_epoch=False)
# schedule
runner = dict(max_epochs=400)
# runtime
checkpoint_config = dict(interval=1, max_keep_ckpts=3, out_dir='')
persistent_workers = True
log_config = dict(
interval=100, hooks=[
dict(type='TextLoggerHook'),
])

4
configs/selfsup/mae/mae_vit-base-p16_8xb512-coslr-800e_in1k.py 100644
View File

@ -0,0 +1,4 @@
_base_ = 'mae_vit-base-16_8xb512-coslr-400e_in1k.py'
# schedule
runner = dict(max_epochs=800)

50
configs/selfsup/mocov2/README.md
View File

@ -14,7 +14,7 @@ Contrastive unsupervised learning has recently shown encouraging progress, e.g.,
## Results and Models
**Back to [model_zoo.md](../../../docs/en/model_zoo.md) to download models.**
**Back to [model_zoo.md](https://github.com/open-mmlab/mmselfsup/blob/master/docs/en/model_zoo.md) to download models.**
In this page, we provide benchmarks as much as possible to evaluate our pre-trained models. If not mentioned, all models were trained on ImageNet1k dataset.
@ -28,19 +28,27 @@ The **Best Layer** indicates that the best results are obtained from which layer
Besides, k=1 to 96 indicates the hyper-parameter of Low-shot SVM.
| Self-Supervised Config | Best Layer | SVM | k=1 | k=2 | k=4 | k=8 | k=16 | k=32 | k=64 | k=96 |
| ---------------------------------------------------------------------------- | ---------- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- |
| [mocov2_resnet50_8xb32-coslr-200e](mocov2_resnet50_8xb32-coslr-200e_in1k.py) | feature5 | 84.04 | 43.14 | 53.29 | 65.34 | 71.03 | 75.42 | 78.48 | 80.88 | 82.23 |
| Self-Supervised Config | Best Layer | SVM | k=1 | k=2 | k=4 | k=8 | k=16 | k=32 | k=64 | k=96 |
| ------------------------------------------------------------------------------------------------------------------------------------------------ | ---------- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- |
| [resnet50_8xb32-coslr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mocov2/mocov2_resnet50_8xb32-coslr-200e_in1k.py) | feature5 | 84.04 | 43.14 | 53.29 | 65.34 | 71.03 | 75.42 | 78.48 | 80.88 | 82.23 |
#### ImageNet Linear Evaluation
The **Feature1 - Feature5** don't have the GlobalAveragePooling, the feature map is pooled to the specific dimensions and then follows a Linear layer to do the classification. Please refer to [resnet50_mhead_8xb32-steplr-90e.py](../../benchmarks/classification/imagenet/resnet50_mhead_8xb32-steplr-90e_in1k.py) for details of config.
The **Feature1 - Feature5** don't have the GlobalAveragePooling, the feature map is pooled to the specific dimensions and then follows a Linear layer to do the classification. Please refer to [resnet50_mhead_8xb32-steplr-90e.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/classification/imagenet/resnet50_mhead_8xb32-steplr-90e_in1k.py) for details of config.
The **AvgPool** result is obtained from Linear Evaluation with GlobalAveragePooling. Please refer to [resnet50_8xb32-steplr-100e_in1k](../../benchmarks/classification/imagenet/resnet50_8xb32-steplr-100e_in1k.py) for details of config.
The **AvgPool** result is obtained from Linear Evaluation with GlobalAveragePooling. Please refer to [resnet50_8xb32-steplr-100e_in1k.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/classification/imagenet/resnet50_8xb32-steplr-100e_in1k.py) for details of config.
| Self-Supervised Config | Feature1 | Feature2 | Feature3 | Feature4 | Feature5 | AvgPool |
| ---------------------------------------------------------------------------- | -------- | -------- | -------- | -------- | -------- | ------- |
| [mocov2_resnet50_8xb32-coslr-200e](mocov2_resnet50_8xb32-coslr-200e_in1k.py) | 15.96 | 34.22 | 45.78 | 61.11 | 66.24 | 67.56 |
| Self-Supervised Config | Feature1 | Feature2 | Feature3 | Feature4 | Feature5 | AvgPool |
| ------------------------------------------------------------------------------------------------------------------------------------------------ | -------- | -------- | -------- | -------- | -------- | ------- |
| [resnet50_8xb32-coslr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mocov2/mocov2_resnet50_8xb32-coslr-200e_in1k.py) | 15.96 | 34.22 | 45.78 | 61.11 | 66.24 | 67.58 |
#### Places205 Linear Evaluation
The **Feature1 - Feature5** don't have the GlobalAveragePooling, the feature map is pooled to the specific dimensions and then follows a Linear layer to do the classification. Please refer to [resnet50_mhead_8xb32-steplr-28e_places205.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/classification/places205/resnet50_mhead_8xb32-steplr-28e_places205.py) for details of config.
| Self-Supervised Config | Feature1 | Feature2 | Feature3 | Feature4 | Feature5 |
| ------------------------------------------------------------------------------------------------------------------------------------------------ | -------- | -------- | -------- | -------- | -------- |
| [resnet50_8xb32-coslr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mocov2/mocov2_resnet50_8xb32-coslr-200e_in1k.py) | 20.92 | 35.72 | 42.62 | 49.79 | 52.25 |
### Detection
@ -48,19 +56,19 @@ The detection benchmarks includes 2 downstream task datasets, **Pascal VOC 2007
#### Pascal VOC 2007 + 2012
Please refer to [faster_rcnn_r50_c4_mstrain_24k_voc0712.py](../../benchmarks/mmdetection/voc0712/faster_rcnn_r50_c4_mstrain_24k_voc0712.py) for details of config.
Please refer to [faster_rcnn_r50_c4_mstrain_24k_voc0712.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/mmdetection/voc0712/faster_rcnn_r50_c4_mstrain_24k_voc0712.py) for details of config.
| Self-Supervised Config | AP50 |
| ---------------------------------------------------------------------------- | ----- |
| [mocov2_resnet50_8xb32-coslr-200e](mocov2_resnet50_8xb32-coslr-200e_in1k.py) | 81.06 |
| Self-Supervised Config | AP50 |
| ------------------------------------------------------------------------------------------------------------------------------------------------ | ----- |
| [resnet50_8xb32-coslr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mocov2/mocov2_resnet50_8xb32-coslr-200e_in1k.py) | 81.06 |
#### COCO2017
Please refer to [mask_rcnn_r50_fpn_mstrain_1x_coco.py](../../benchmarks/mmdetection/coco/mask_rcnn_r50_fpn_mstrain_1x_coco.py) for details of config.
Please refer to [mask_rcnn_r50_fpn_mstrain_1x_coco.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/mmdetection/coco/mask_rcnn_r50_fpn_mstrain_1x_coco.py) for details of config.
| Self-Supervised Config | mAP(Box) | AP50(Box) | AP75(Box) | mAP(Mask) | AP50(Mask) | AP75(Mask) |
| ---------------------------------------------------------------------------- | -------- | --------- | --------- | --------- | ---------- | ---------- |
| [mocov2_resnet50_8xb32-coslr-200e](mocov2_resnet50_8xb32-coslr-200e_in1k.py) | 40.2 | 59.7 | 44.2 | 36.1 | 56.7 | 38.8 |
| Self-Supervised Config | mAP(Box) | AP50(Box) | AP75(Box) | mAP(Mask) | AP50(Mask) | AP75(Mask) |
| ------------------------------------------------------------------------------------------------------------------------------------------------ | -------- | --------- | --------- | --------- | ---------- | ---------- |
| [resnet50_8xb32-coslr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mocov2/mocov2_resnet50_8xb32-coslr-200e_in1k.py) | 40.2 | 59.7 | 44.2 | 36.1 | 56.7 | 38.8 |
### Segmentation
@ -68,11 +76,11 @@ The segmentation benchmarks includes 2 downstream task datasets, **Cityscapes**
#### Pascal VOC 2012 + Aug
Please refer to [fcn_r50-d8_512x512_20k_voc12aug.py](../../benchmarks/mmsegmentation/voc12aug/fcn_r50-d8_512x512_20k_voc12aug.py) for details of config.
Please refer to [fcn_r50-d8_512x512_20k_voc12aug.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/mmsegmentation/voc12aug/fcn_r50-d8_512x512_20k_voc12aug.py) for details of config.
| Self-Supervised Config | mIOU |
| ---------------------------------------------------------------------------- | ----- |
| [mocov2_resnet50_8xb32-coslr-200e](mocov2_resnet50_8xb32-coslr-200e_in1k.py) | 67.55 |
| Self-Supervised Config | mIOU |
| ------------------------------------------------------------------------------------------------------------------------------------------------ | ----- |
| [resnet50_8xb32-coslr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mocov2/mocov2_resnet50_8xb32-coslr-200e_in1k.py) | 67.55 |
## Citation

4
configs/selfsup/mocov2/metafile.yml
View File

@ -24,6 +24,6 @@ Models:
- Task: Self-Supervised Image Classification
Dataset: ImageNet-1k
Metrics:
Top 1 Accuracy: 67.56
Top 1 Accuracy: 67.58
Config: configs/selfsup/mocov2/mocov2_resnet50_8xb32-coslr-200e_in1k.py
Weights: https://download.openmmlab.com/mmselfsup/moco/mocov2_resnet50_8xb32-coslr-200e_in1k_20211213-7ce8f840.pth
Weights: https://download.openmmlab.com/mmselfsup/moco/mocov2_resnet50_8xb32-coslr-200e_in1k_20220225-89e03af4.pth

10
configs/selfsup/mocov3/README.md
View File

@ -14,7 +14,7 @@ This paper does not describe a novel method. Instead, it studies a straightforwa
## Results and Models
**Back to [model_zoo.md](../../../docs/en/model_zoo.md) to download models.**
**Back to [model_zoo.md](https://github.com/open-mmlab/mmselfsup/blob/master/docs/en/model_zoo.md) to download models.**
In this page, we provide benchmarks as much as possible to evaluate our pre-trained models. If not mentioned, all models were trained on ImageNet1k dataset.
@ -24,11 +24,11 @@ The classification benchmarks includes 4 downstream task datasets, **VOC**, **Im
#### ImageNet Linear Evaluation
The **Linear Evaluation** result is obtained by training a linear head upon the pre-trained backbone. Please refer to [vit-small-p16_8xb128-coslr-90e_in1k](../../benchmarks/classification/imagenet/vit-small-p16_8xb128-coslr-90e_in1k.py) for details of config.
The **Linear Evaluation** result is obtained by training a linear head upon the pre-trained backbone. Please refer to [vit-small-p16_8xb128-coslr-90e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/classification/imagenet/vit-small-p16_8xb128-coslr-90e_in1k.py) for details of config.
| Self-Supervised Config | Linear Evaluation |
| ----------------------------------------------------------------------------------------------------------------- | ----------------- |
| [mocov3_vit-small-p16_32xb128-fp16-coslr-300e_in1k-224](mocov3_vit-small-p16_32xb128-fp16-coslr-300e_in1k-224.py) | 73.07 |
| Self-Supervised Config | Linear Evaluation |
| ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----------------- |
| [vit-small-p16_32xb128-fp16-coslr-300e_in1k-224](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mocov3/mocov3_vit-small-p16_32xb128-fp16-coslr-300e_in1k-224.py) | 73.19 |
## Citation

4
configs/selfsup/mocov3/metafile.yml
View File

@ -23,6 +23,6 @@ Models:
- Task: Self-Supervised Image Classification
Dataset: ImageNet-1k
Metrics:
Top 1 Accuracy: 73.07
Top 1 Accuracy: 73.19
Config: configs/selfsup/mocov3/mocov3_vit-small-p16_32xb128-fp16-coslr-300e_in1k-224.py
Weights: https://download.openmmlab.com/mmselfsup/moco/mocov3_vit-small-p16_32xb128-fp16-coslr-300e_in1k-224_20220127-e9332db2.pth
Weights: https://download.openmmlab.com/mmselfsup/moco/mocov3_vit-small-p16_32xb128-fp16-coslr-300e_in1k-224_20220225-e31238dd.pth

50
configs/selfsup/npid/README.md
View File

@ -18,7 +18,7 @@ Our method is also remarkable for consistently improving test performance with m
## Results and Models
**Back to [model_zoo.md](../../../docs/en/model_zoo.md) to download models.**
**Back to [model_zoo.md](https://github.com/open-mmlab/mmselfsup/blob/master/docs/en/model_zoo.md) to download models.**
In this page, we provide benchmarks as much as possible to evaluate our pre-trained models. If not mentioned, all models were trained on ImageNet1k dataset.
@ -32,19 +32,27 @@ The **Best Layer** indicates that the best results are obtained from which layer
Besides, k=1 to 96 indicates the hyper-parameter of Low-shot SVM.
| Self-Supervised Config | Best Layer | SVM | k=1 | k=2 | k=4 | k=8 | k=16 | k=32 | k=64 | k=96 |
| --------------------------------------------------------------------- | ---------- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- |
| [resnet50_8xb32-steplr-200e](npid_resnet50_8xb32-steplr-200e_in1k.py) | feature5 | 76.75 | 26.96 | 35.37 | 44.48 | 53.89 | 60.39 | 66.41 | 71.48 | 73.39 |
| Self-Supervised Config | Best Layer | SVM | k=1 | k=2 | k=4 | k=8 | k=16 | k=32 | k=64 | k=96 |
| ---------------------------------------------------------------------------------------------------------------------------------------------- | ---------- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- |
| [resnet50_8xb32-steplr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/npid/npid_resnet50_8xb32-steplr-200e_in1k.py) | feature5 | 76.75 | 26.96 | 35.37 | 44.48 | 53.89 | 60.39 | 66.41 | 71.48 | 73.39 |
#### ImageNet Linear Evaluation
The **Feature1 - Feature5** don't have the GlobalAveragePooling, the feature map is pooled to the specific dimensions and then follows a Linear layer to do the classification. Please refer to [resnet50_mhead_8xb32-steplr-90e.py](../../benchmarks/classification/imagenet/resnet50_mhead_8xb32-steplr-90e_in1k.py) for details of config.
The **Feature1 - Feature5** don't have the GlobalAveragePooling, the feature map is pooled to the specific dimensions and then follows a Linear layer to do the classification. Please refer to [resnet50_mhead_8xb32-steplr-90e.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/classification/imagenet/resnet50_mhead_8xb32-steplr-90e_in1k.py) for details of config.
The **AvgPool** result is obtained from Linear Evaluation with GlobalAveragePooling. Please refer to [resnet50_8xb32-steplr-100e_in1k](../../benchmarks/classification/imagenet/resnet50_8xb32-steplr-100e_in1k.py) for details of config.
The **AvgPool** result is obtained from Linear Evaluation with GlobalAveragePooling. Please refer to [resnet50_8xb32-steplr-100e_in1k.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/classification/imagenet/resnet50_8xb32-steplr-100e_in1k.py) for details of config.
| Self-Supervised Config | Feature1 | Feature2 | Feature3 | Feature4 | Feature5 | AvgPool |
| --------------------------------------------------------------------- | -------- | -------- | -------- | -------- | -------- | ------- |
| [resnet50_8xb32-steplr-200e](npid_resnet50_8xb32-steplr-200e_in1k.py) | 14.68 | 31.98 | 42.85 | 56.95 | 58.41 | 58.16 |
| Self-Supervised Config | Feature1 | Feature2 | Feature3 | Feature4 | Feature5 | AvgPool |
| ---------------------------------------------------------------------------------------------------------------------------------------------- | -------- | -------- | -------- | -------- | -------- | ------- |
| [resnet50_8xb32-steplr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/npid/npid_resnet50_8xb32-steplr-200e_in1k.py) | 14.68 | 31.98 | 42.85 | 56.95 | 58.41 | 57.97 |
#### Places205 Linear Evaluation
The **Feature1 - Feature5** don't have the GlobalAveragePooling, the feature map is pooled to the specific dimensions and then follows a Linear layer to do the classification. Please refer to [resnet50_mhead_8xb32-steplr-28e_places205.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/classification/places205/resnet50_mhead_8xb32-steplr-28e_places205.py) for details of config.
| Self-Supervised Config | Feature1 | Feature2 | Feature3 | Feature4 | Feature5 |
| ---------------------------------------------------------------------------------------------------------------------------------------------- | -------- | -------- | -------- | -------- | -------- |
| [resnet50_8xb32-steplr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/npid/npid_resnet50_8xb32-steplr-200e_in1k.py) | 19.98 | 34.86 | 41.59 | 48.43 | 48.71 |
### Detection
@ -52,19 +60,19 @@ The detection benchmarks includes 2 downstream task datasets, **Pascal VOC 2007
#### Pascal VOC 2007 + 2012
Please refer to [faster_rcnn_r50_c4_mstrain_24k_voc0712.py](../../benchmarks/mmdetection/voc0712/faster_rcnn_r50_c4_mstrain_24k_voc0712.py) for details of config.
Please refer to [faster_rcnn_r50_c4_mstrain_24k_voc0712.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/mmdetection/voc0712/faster_rcnn_r50_c4_mstrain_24k_voc0712.py) for details of config.
| Self-Supervised Config | AP50 |
| --------------------------------------------------------------------- | ----- |
| [resnet50_8xb32-steplr-200e](npid_resnet50_8xb32-steplr-200e_in1k.py) | 79.52 |
| Self-Supervised Config | AP50 |
| ---------------------------------------------------------------------------------------------------------------------------------------------- | ----- |
| [resnet50_8xb32-steplr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/npid/npid_resnet50_8xb32-steplr-200e_in1k.py) | 79.52 |
#### COCO2017
Please refer to [mask_rcnn_r50_fpn_mstrain_1x_coco.py](../../benchmarks/mmdetection/coco/mask_rcnn_r50_fpn_mstrain_1x_coco.py) for details of config.
Please refer to [mask_rcnn_r50_fpn_mstrain_1x_coco.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/mmdetection/coco/mask_rcnn_r50_fpn_mstrain_1x_coco.py) for details of config.
| Self-Supervised Config | mAP(Box) | AP50(Box) | AP75(Box) | mAP(Mask) | AP50(Mask) | AP75(Mask) |
| --------------------------------------------------------------------- | -------- | --------- | --------- | --------- | ---------- | ---------- |
| [resnet50_8xb32-steplr-200e](npid_resnet50_8xb32-steplr-200e_in1k.py) | 38.5 | 57.7 | 42.0 | 34.6 | 54.8 | 37.1 |
| Self-Supervised Config | mAP(Box) | AP50(Box) | AP75(Box) | mAP(Mask) | AP50(Mask) | AP75(Mask) |
| ---------------------------------------------------------------------------------------------------------------------------------------------- | -------- | --------- | --------- | --------- | ---------- | ---------- |
| [resnet50_8xb32-steplr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/npid/npid_resnet50_8xb32-steplr-200e_in1k.py) | 38.5 | 57.7 | 42.0 | 34.6 | 54.8 | 37.1 |
### Segmentation
@ -72,11 +80,11 @@ The segmentation benchmarks includes 2 downstream task datasets, **Cityscapes**
#### Pascal VOC 2012 + Aug
Please refer to [fcn_r50-d8_512x512_20k_voc12aug.py](../../benchmarks/mmsegmentation/voc12aug/fcn_r50-d8_512x512_20k_voc12aug.py) for details of config.
Please refer to [fcn_r50-d8_512x512_20k_voc12aug.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/mmsegmentation/voc12aug/fcn_r50-d8_512x512_20k_voc12aug.py) for details of config.
| Self-Supervised Config | mIOU |
| --------------------------------------------------------------------- | ----- |
| [resnet50_8xb32-steplr-200e](npid_resnet50_8xb32-steplr-200e_in1k.py) | 65.45 |
| Self-Supervised Config | mIOU |
| ---------------------------------------------------------------------------------------------------------------------------------------------- | ----- |
| [resnet50_8xb32-steplr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/npid/npid_resnet50_8xb32-steplr-200e_in1k.py) | 65.45 |
## Citation

4
configs/selfsup/npid/metafile.yml
View File

@ -24,6 +24,6 @@ Models:
- Task: Self-Supervised Image Classification
Dataset: ImageNet-1k
Metrics:
Top 1 Accuracy: 58.16
Top 1 Accuracy: 57.97
Config: configs/selfsup/npid/npid_resnet50_8xb32-steplr-200e_in1k.py
Weights: https://download.openmmlab.com/mmselfsup/npid/npid_resnet50_8xb32-steplr-200e_in1k_20211213-b5fec6df.pth
Weights: https://download.openmmlab.com/mmselfsup/npid/npid_resnet50_8xb32-steplr-200e_in1k_20220225-5fbbda2a.pth

26
configs/selfsup/odc/README.md
View File

@ -14,7 +14,7 @@ Joint clustering and feature learning methods have shown remarkable performance
## Results and Models
**Back to [model_zoo.md](../../../docs/en/model_zoo.md) to download models.**
**Back to [model_zoo.md](https://github.com/open-mmlab/mmselfsup/blob/master/docs/en/model_zoo.md) to download models.**
In this page, we provide benchmarks as much as possible to evaluate our pre-trained models. If not mentioned, all models were trained on ImageNet1k dataset.
@ -28,19 +28,27 @@ The **Best Layer** indicates that the best results are obtained from which layer
Besides, k=1 to 96 indicates the hyper-parameter of Low-shot SVM.
| Self-Supervised Config | Best Layer | SVM | k=1 | k=2 | k=4 | k=8 | k=16 | k=32 | k=64 | k=96 |
| -------------------------------------------------------------------- | ---------- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- |
| [resnet50_8xb64-steplr-440e](odc_resnet50_8xb64-steplr-440e_in1k.py) | feature5 | 78.42 | 32.42 | 40.27 | 49.95 | 59.96 | 65.71 | 69.99 | 73.64 | 75.13 |
| Self-Supervised Config | Best Layer | SVM | k=1 | k=2 | k=4 | k=8 | k=16 | k=32 | k=64 | k=96 |
| -------------------------------------------------------------------------------------------------------------------------------------------- | ---------- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- |
| [resnet50_8xb64-steplr-440e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/odc/odc_resnet50_8xb64-steplr-440e_in1k.py) | feature5 | 78.42 | 32.42 | 40.27 | 49.95 | 59.96 | 65.71 | 69.99 | 73.64 | 75.13 |
#### ImageNet Linear Evaluation
The **Feature1 - Feature5** don't have the GlobalAveragePooling, the feature map is pooled to the specific dimensions and then follows a Linear layer to do the classification. Please refer to [resnet50_mhead_8xb32-steplr-90e.py](../../benchmarks/classification/imagenet/resnet50_mhead_8xb32-steplr-90e_in1k.py) for details of config.
The **Feature1 - Feature5** don't have the GlobalAveragePooling, the feature map is pooled to the specific dimensions and then follows a Linear layer to do the classification. Please refer to [resnet50_mhead_8xb32-steplr-90e.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/classification/imagenet/resnet50_mhead_8xb32-steplr-90e_in1k.py) for details of config.
The **AvgPool** result is obtained from Linear Evaluation with GlobalAveragePooling. Please refer to [resnet50_8xb32-steplr-100e_in1k](../../benchmarks/classification/imagenet/resnet50_8xb32-steplr-100e_in1k.py) for details of config.
The **AvgPool** result is obtained from Linear Evaluation with GlobalAveragePooling. Please refer to [resnet50_8xb32-steplr-100e_in1k.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/classification/imagenet/resnet50_8xb32-steplr-100e_in1k.py) for details of config.
| Self-Supervised Config | Feature1 | Feature2 | Feature3 | Feature4 | Feature5 | AvgPool |
| -------------------------------------------------------------------- | -------- | -------- | -------- | -------- | -------- | ------- |
| [resnet50_8xb64-steplr-440e](odc_resnet50_8xb64-steplr-440e_in1k.py) | 14.76 | 31.82 | 42.44 | 55.76 | 57.70 | 53.42 |
| Self-Supervised Config | Feature1 | Feature2 | Feature3 | Feature4 | Feature5 | AvgPool |
| -------------------------------------------------------------------------------------------------------------------------------------------- | -------- | -------- | -------- | -------- | -------- | ------- |
| [resnet50_8xb64-steplr-440e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/odc/odc_resnet50_8xb64-steplr-440e_in1k.py) | 14.76 | 31.82 | 42.44 | 55.76 | 57.70 | 53.42 |
#### Places205 Linear Evaluation
The **Feature1 - Feature5** don't have the GlobalAveragePooling, the feature map is pooled to the specific dimensions and then follows a Linear layer to do the classification. Please refer to [resnet50_mhead_8xb32-steplr-28e_places205.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/classification/places205/resnet50_mhead_8xb32-steplr-28e_places205.py) for details of config.
| Self-Supervised Config | Feature1 | Feature2 | Feature3 | Feature4 | Feature5 |
| -------------------------------------------------------------------------------------------------------------------------------------------- | -------- | -------- | -------- | -------- | -------- |
| [resnet50_8xb64-steplr-440e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/odc/odc_resnet50_8xb64-steplr-440e_in1k.py) | 19.28 | 34.09 | 40.90 | 47.04 | 48.35 |
## Citation

4
configs/selfsup/odc/metafile.yml
View File

@ -23,6 +23,6 @@ Models:
- Task: Self-Supervised Image Classification
Dataset: ImageNet-1k
Metrics:
Top 1 Accuracy: 53.42
Top 1 Accuracy: 53.43
Config: configs/selfsup/odc/odc_resnet50_8xb64-steplr-440e_in1k.py
Weights: https://download.openmmlab.com/mmselfsup/odc/odc_resnet50_8xb64-steplr-440e_in1k-5af5dd0c.pth
Weights: https://download.openmmlab.com/mmselfsup/odc/odc_resnet50_8xb64-steplr-440e_in1k_20220225-a755d9c0.pth

50
configs/selfsup/relative_loc/README.md
View File

@ -14,7 +14,7 @@ This work explores the use of spatial context as a source of free and plentiful
## Results and Models
**Back to [model_zoo.md](../../../docs/en/model_zoo.md) to download models.**
**Back to [model_zoo.md](https://github.com/open-mmlab/mmselfsup/blob/master/docs/en/model_zoo.md) to download models.**
In this page, we provide benchmarks as much as possible to evaluate our pre-trained models. If not mentioned, all models were trained on ImageNet1k dataset.
@ -28,19 +28,27 @@ The **Best Layer** indicates that the best results are obtained from which layer
Besides, k=1 to 96 indicates the hyper-parameter of Low-shot SVM.
| Self-Supervised Config | Best Layer | SVM | k=1 | k=2 | k=4 | k=8 | k=16 | k=32 | k=64 | k=96 |
| --------------------------------------------------------------------------- | ---------- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- |
| [resnet50_8xb64-steplr-70e](relative-loc_resnet50_8xb64-steplr-70e_in1k.py) | feature4 | 65.52 | 20.36 | 23.12 | 30.66 | 37.02 | 42.55 | 50.00 | 55.58 | 59.28 |
| Self-Supervised Config | Best Layer | SVM | k=1 | k=2 | k=4 | k=8 | k=16 | k=32 | k=64 | k=96 |
| ------------------------------------------------------------------------------------------------------------------------------------------------------------ | ---------- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- |
| [resnet50_8xb64-steplr-70e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/relative_loc/relative-loc_resnet50_8xb64-steplr-70e_in1k.py) | feature4 | 65.52 | 20.36 | 23.12 | 30.66 | 37.02 | 42.55 | 50.00 | 55.58 | 59.28 |
#### ImageNet Linear Evaluation
The **Feature1 - Feature5** don't have the GlobalAveragePooling, the feature map is pooled to the specific dimensions and then follows a Linear layer to do the classification. Please refer to [resnet50_mhead_8xb32-steplr-90e.py](../../benchmarks/classification/imagenet/resnet50_mhead_8xb32-steplr-90e_in1k.py) for details of config.
The **Feature1 - Feature5** don't have the GlobalAveragePooling, the feature map is pooled to the specific dimensions and then follows a Linear layer to do the classification. Please refer to [resnet50_mhead_8xb32-steplr-90e.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/classification/imagenet/resnet50_mhead_8xb32-steplr-90e_in1k.py) for details of config.
The **AvgPool** result is obtained from Linear Evaluation with GlobalAveragePooling. Please refer to [resnet50_8xb32-steplr-100e_in1k](../../benchmarks/classification/imagenet/resnet50_8xb32-steplr-100e_in1k.py) for details of config.
The **AvgPool** result is obtained from Linear Evaluation with GlobalAveragePooling. Please refer to [resnet50_8xb32-steplr-100e_in1k.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/classification/imagenet/resnet50_8xb32-steplr-100e_in1k.py) for details of config.
| Self-Supervised Config | Feature1 | Feature2 | Feature3 | Feature4 | Feature5 | AvgPool |
| --------------------------------------------------------------------------- | -------- | -------- | -------- | -------- | -------- | ------- |
| [resnet50_8xb64-steplr-70e](relative-loc_resnet50_8xb64-steplr-70e_in1k.py) | 15.11 | 30.47 | 42.83 | 51.20 | 40.96 | 39.65 |
| Self-Supervised Config | Feature1 | Feature2 | Feature3 | Feature4 | Feature5 | AvgPool |
| ------------------------------------------------------------------------------------------------------------------------------------------------------------ | -------- | -------- | -------- | -------- | -------- | ------- |
| [resnet50_8xb64-steplr-70e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/relative_loc/relative-loc_resnet50_8xb64-steplr-70e_in1k.py) | 15.11 | 30.47 | 42.83 | 51.20 | 40.96 | 39.65 |
#### Places205 Linear Evaluation
The **Feature1 - Feature5** don't have the GlobalAveragePooling, the feature map is pooled to the specific dimensions and then follows a Linear layer to do the classification. Please refer to [resnet50_mhead_8xb32-steplr-28e_places205.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/classification/places205/resnet50_mhead_8xb32-steplr-28e_places205.py) for details of config.
| Self-Supervised Config | Feature1 | Feature2 | Feature3 | Feature4 | Feature5 |
| ------------------------------------------------------------------------------------------------------------------------------------------------------------ | -------- | -------- | -------- | -------- | -------- |
| [resnet50_8xb64-steplr-70e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/relative_loc/relative-loc_resnet50_8xb64-steplr-70e_in1k.py) | 20.69 | 34.72 | 43.01 | 45.97 | 41.96 |
### Detection
@ -48,19 +56,19 @@ The detection benchmarks includes 2 downstream task datasets, **Pascal VOC 2007
#### Pascal VOC 2007 + 2012
Please refer to [faster_rcnn_r50_c4_mstrain_24k_voc0712.py](../../benchmarks/mmdetection/voc0712/faster_rcnn_r50_c4_mstrain_24k_voc0712.py) for details of config.
Please refer to [faster_rcnn_r50_c4_mstrain_24k_voc0712.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/mmdetection/voc0712/faster_rcnn_r50_c4_mstrain_24k_voc0712.py) for details of config.
| Self-Supervised Config | AP50 |
| --------------------------------------------------------------------------- | ----- |
| [resnet50_8xb64-steplr-70e](relative-loc_resnet50_8xb64-steplr-70e_in1k.py) | 79.70 |
| Self-Supervised Config | AP50 |
| ------------------------------------------------------------------------------------------------------------------------------------------------------------ | ----- |
| [resnet50_8xb64-steplr-70e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/relative_loc/relative-loc_resnet50_8xb64-steplr-70e_in1k.py) | 79.70 |
#### COCO2017
Please refer to [mask_rcnn_r50_fpn_mstrain_1x_coco.py](../../benchmarks/mmdetection/coco/mask_rcnn_r50_fpn_mstrain_1x_coco.py) for details of config.
Please refer to [mask_rcnn_r50_fpn_mstrain_1x_coco.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/mmdetection/coco/mask_rcnn_r50_fpn_mstrain_1x_coco.py) for details of config.
| Self-Supervised Config | mAP(Box) | AP50(Box) | AP75(Box) | mAP(Mask) | AP50(Mask) | AP75(Mask) |
| --------------------------------------------------------------------------- | -------- | --------- | --------- | --------- | ---------- | ---------- |
| [resnet50_8xb64-steplr-70e](relative-loc_resnet50_8xb64-steplr-70e_in1k.py) | 37.5 | 56.2 | 41.3 | 33.7 | 53.3 | 36.1 |
| Self-Supervised Config | mAP(Box) | AP50(Box) | AP75(Box) | mAP(Mask) | AP50(Mask) | AP75(Mask) |
| -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | -------- | --------- | --------- | --------- | ---------- | ---------- |
| [resnet50_8xb64-steplr-70e]([relative-loc_resnet50_8xb64-steplr-70e_in1k.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/relative_loc/relative-loc_resnet50_8xb64-steplr-70e_in1k.py)) | 37.5 | 56.2 | 41.3 | 33.7 | 53.3 | 36.1 |
### Segmentation
@ -68,11 +76,11 @@ The segmentation benchmarks includes 2 downstream task datasets, **Cityscapes**
#### Pascal VOC 2012 + Aug
Please refer to [fcn_r50-d8_512x512_20k_voc12aug.py](../../benchmarks/mmsegmentation/voc12aug/fcn_r50-d8_512x512_20k_voc12aug.py) for details of config.
Please refer to [fcn_r50-d8_512x512_20k_voc12aug.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/mmsegmentation/voc12aug/fcn_r50-d8_512x512_20k_voc12aug.py) for details of config.
| Self-Supervised Config | mIOU |
| --------------------------------------------------------------------------- | ----- |
| [resnet50_8xb64-steplr-70e](relative-loc_resnet50_8xb64-steplr-70e_in1k.py) | 63.49 |
| Self-Supervised Config | mIOU |
| ------------------------------------------------------------------------------------------------------------------------------------------------------------ | ----- |
| [resnet50_8xb64-steplr-70e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/relative_loc/relative-loc_resnet50_8xb64-steplr-70e_in1k.py) | 63.49 |
## Citation

4
configs/selfsup/relative_loc/metafile.yml
View File

@ -23,6 +23,6 @@ Models:
- Task: Self-Supervised Image Classification
Dataset: ImageNet-1k
Metrics:
Top 1 Accuracy: 39.65
Top 1 Accuracy: 38.78
Config: configs/selfsup/relative_loc/relative-loc_resnet50_8xb64-steplr-70e_in1k.py
Weights: https://download.openmmlab.com/mmselfsup/relative_loc/relative-loc_resnet50_8xb64-steplr-70e_in1k_20211213-cdd3162f.pth
Weights: https://download.openmmlab.com/mmselfsup/relative_loc/relative-loc_resnet50_8xb64-steplr-70e_in1k_20220225-84784688.pth

50
configs/selfsup/rotation_pred/README.md
View File

@ -14,7 +14,7 @@ Over the last years, deep convolutional neural networks (ConvNets) have transfor
## Results and Models
**Back to [model_zoo.md](../../../docs/en/model_zoo.md) to download models.**
**Back to [model_zoo.md](https://github.com/open-mmlab/mmselfsup/blob/master/docs/en/model_zoo.md) to download models.**
In this page, we provide benchmarks as much as possible to evaluate our pre-trained models. If not mentioned, all models were trained on ImageNet1k dataset.
@ -28,19 +28,27 @@ The **Best Layer** indicates that the best results are obtained from which layer
Besides, k=1 to 96 indicates the hyper-parameter of Low-shot SVM.
| Self-Supervised Config | Best Layer | SVM | k=1 | k=2 | k=4 | k=8 | k=16 | k=32 | k=64 | k=96 |
| ---------------------------------------------------------------------------- | ---------- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- |
| [resnet50_8xb16-steplr-70e](rotation-pred_resnet50_8xb16-steplr-70e_in1k.py) | feature4 | 67.70 | 20.60 | 24.35 | 31.41 | 39.17 | 46.56 | 53.37 | 59.14 | 62.42 |
| Self-Supervised Config | Best Layer | SVM | k=1 | k=2 | k=4 | k=8 | k=16 | k=32 | k=64 | k=96 |
| -------------------------------------------------------------------------------------------------------------------------------------------------------------- | ---------- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- |
| [resnet50_8xb16-steplr-70e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/rotation_pred/rotation-pred_resnet50_8xb16-steplr-70e_in1k.py) | feature4 | 67.70 | 20.60 | 24.35 | 31.41 | 39.17 | 46.56 | 53.37 | 59.14 | 62.42 |
#### ImageNet Linear Evaluation
The **Feature1 - Feature5** don't have the GlobalAveragePooling, the feature map is pooled to the specific dimensions and then follows a Linear layer to do the classification. Please refer to [resnet50_mhead_8xb32-steplr-90e.py](../../benchmarks/classification/imagenet/resnet50_mhead_8xb32-steplr-90e_in1k.py) for details of config.
The **Feature1 - Feature5** don't have the GlobalAveragePooling, the feature map is pooled to the specific dimensions and then follows a Linear layer to do the classification. Please refer to [resnet50_mhead_8xb32-steplr-90e.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/classification/imagenet/resnet50_mhead_8xb32-steplr-90e_in1k.py) for details of config.
The **AvgPool** result is obtained from Linear Evaluation with GlobalAveragePooling. Please refer to [file name]() for details of config.
The **AvgPool** result is obtained from Linear Evaluation with GlobalAveragePooling. Please refer to [resnet50_8xb32-steplr-100e_in1k.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/classification/imagenet/resnet50_8xb32-steplr-100e_in1k.py) for details of config.
| Self-Supervised Config | Feature1 | Feature2 | Feature3 | Feature4 | Feature5 | AvgPool |
| ---------------------------------------------------------------------------- | -------- | -------- | -------- | -------- | -------- | ------- |
| [resnet50_8xb16-steplr-70e](rotation-pred_resnet50_8xb16-steplr-70e_in1k.py) | 12.15 | 31.99 | 44.57 | 54.20 | 45.94 | 44.35 |
| Self-Supervised Config | Feature1 | Feature2 | Feature3 | Feature4 | Feature5 | AvgPool |
| -------------------------------------------------------------------------------------------------------------------------------------------------------------- | -------- | -------- | -------- | -------- | -------- | ------- |
| [resnet50_8xb16-steplr-70e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/rotation_pred/rotation-pred_resnet50_8xb16-steplr-70e_in1k.py) | 12.15 | 31.99 | 44.57 | 54.20 | 45.94 | 48.12 |
#### Places205 Linear Evaluation
The **Feature1 - Feature5** don't have the GlobalAveragePooling, the feature map is pooled to the specific dimensions and then follows a Linear layer to do the classification. Please refer to [resnet50_mhead_8xb32-steplr-28e_places205.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/classification/places205/resnet50_mhead_8xb32-steplr-28e_places205.py) for details of config.
| Self-Supervised Config | Feature1 | Feature2 | Feature3 | Feature4 | Feature5 |
| -------------------------------------------------------------------------------------------------------------------------------------------------------------- | -------- | -------- | -------- | -------- | -------- |
| [resnet50_8xb16-steplr-70e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/rotation_pred/rotation-pred_resnet50_8xb16-steplr-70e_in1k.py) | 18.94 | 34.72 | 44.53 | 46.30 | 44.12 |
### Detection
@ -48,19 +56,19 @@ The detection benchmarks includes 2 downstream task datasets, **Pascal VOC 2007
#### Pascal VOC 2007 + 2012
Please refer to [faster_rcnn_r50_c4_mstrain_24k_voc0712.py](../../benchmarks/mmdetection/voc0712/faster_rcnn_r50_c4_mstrain_24k_voc0712.py) for details of config.
Please refer to [faster_rcnn_r50_c4_mstrain_24k_voc0712.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/mmdetection/voc0712/faster_rcnn_r50_c4_mstrain_24k_voc0712.py) for details of config.
| Self-Supervised Config | AP50 |
| ---------------------------------------------------------------------------- | ----- |
| [resnet50_8xb16-steplr-70e](rotation-pred_resnet50_8xb16-steplr-70e_in1k.py) | 79.67 |
| Self-Supervised Config | AP50 |
| ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----- |
| [resnet50_8xb16-steplr-70e]([rotation-pred_resnet50_8xb16-steplr-70e_in1k.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/rotation_pred/rotation-pred_resnet50_8xb16-steplr-70e_in1k.py)) | 79.67 |
#### COCO2017
Please refer to [mask_rcnn_r50_fpn_mstrain_1x_coco.py](../../benchmarks/mmdetection/coco/mask_rcnn_r50_fpn_mstrain_1x_coco.py) for details of config.
Please refer to [mask_rcnn_r50_fpn_mstrain_1x_coco.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/mmdetection/coco/mask_rcnn_r50_fpn_mstrain_1x_coco.py) for details of config.
| Self-Supervised Config | mAP(Box) | AP50(Box) | AP75(Box) | mAP(Mask) | AP50(Mask) | AP75(Mask) |
| ---------------------------------------------------------------------------- | -------- | --------- | --------- | --------- | ---------- | ---------- |
| [resnet50_8xb16-steplr-70e](rotation-pred_resnet50_8xb16-steplr-70e_in1k.py) | 37.9 | 56.5 | 41.5 | 34.2 | 53.9 | 36.7 |
| Self-Supervised Config | mAP(Box) | AP50(Box) | AP75(Box) | mAP(Mask) | AP50(Mask) | AP75(Mask) |
| ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | -------- | --------- | --------- | --------- | ---------- | ---------- |
| [resnet50_8xb16-steplr-70e]([rotation-pred_resnet50_8xb16-steplr-70e_in1k.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/rotation_pred/rotation-pred_resnet50_8xb16-steplr-70e_in1k.py)) | 37.9 | 56.5 | 41.5 | 34.2 | 53.9 | 36.7 |
### Segmentation
@ -68,11 +76,11 @@ The segmentation benchmarks includes 2 downstream task datasets, **Cityscapes**
#### Pascal VOC 2012 + Aug
Please refer to [fcn_r50-d8_512x512_20k_voc12aug.py](../../benchmarks/mmsegmentation/voc12aug/fcn_r50-d8_512x512_20k_voc12aug.py) for details of config.
Please refer to [fcn_r50-d8_512x512_20k_voc12aug.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/mmsegmentation/voc12aug/fcn_r50-d8_512x512_20k_voc12aug.py) for details of config.
| Self-Supervised Config | mIOU |
| ---------------------------------------------------------------------------- | ----- |
| [resnet50_8xb16-steplr-70e](rotation-pred_resnet50_8xb16-steplr-70e_in1k.py) | 64.31 |
| Self-Supervised Config | mIOU |
| ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----- |
| [resnet50_8xb16-steplr-70e]([rotation-pred_resnet50_8xb16-steplr-70e_in1k.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/rotation_pred/rotation-pred_resnet50_8xb16-steplr-70e_in1k.py)) | 64.31 |
## Citation

4
configs/selfsup/rotation_pred/metafile.yml
View File

@ -23,6 +23,6 @@ Models:
- Task: Self-Supervised Image Classification
Dataset: ImageNet-1k
Metrics:
Top 1 Accuracy: 39.65
Top 1 Accuracy: 48.12
Config: configs/selfsup/rotation_pred/rotation-pred_resnet50_8xb16-steplr-70e_in1k.py
Weights: https://download.openmmlab.com/mmselfsup/rotation_pred/rotation-pred_resnet50_8xb16-steplr-70e_in1k_20211213-513972ac.pth
Weights: https://download.openmmlab.com/mmselfsup/rotation_pred/rotation-pred_resnet50_8xb16-steplr-70e_in1k_20220225-5b9f06a0.pth

50
configs/selfsup/simclr/README.md
View File

@ -14,7 +14,7 @@ This paper presents SimCLR: a simple framework for contrastive learning of visua
## Results and Models
[Back to model_zoo.md](../../../docs/en/model_zoo.md)
**Back to [model_zoo.md](https://github.com/open-mmlab/mmselfsup/blob/master/docs/en/model_zoo.md) to download models.**
In this page, we provide benchmarks as much as possible to evaluate our pre-trained models. If not mentioned, all models were trained on ImageNet1k dataset.
@ -28,19 +28,27 @@ The **Best Layer** indicates that the best results are obtained from which layer
Besides, k=1 to 96 indicates the hyper-parameter of Low-shot SVM.
| Self-Supervised Config | Best Layer | SVM | k=1 | k=2 | k=4 | k=8 | k=16 | k=32 | k=64 | k=96 |
| --------------------------------------------------------------------- | ---------- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ---- |
| [resnet50_8xb32-coslr-200e](simclr_resnet50_8xb32-coslr-200e_in1k.py) | feature5 | 79.98 | 35.02 | 42.79 | 54.87 | 61.91 | 67.38 | 71.88 | 75.56 | 77.4 |
| Self-Supervised Config | Best Layer | SVM | k=1 | k=2 | k=4 | k=8 | k=16 | k=32 | k=64 | k=96 |
| ------------------------------------------------------------------------------------------------------------------------------------------------ | ---------- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ---- |
| [resnet50_8xb32-coslr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simclr/simclr_resnet50_8xb32-coslr-200e_in1k.py) | feature5 | 79.98 | 35.02 | 42.79 | 54.87 | 61.91 | 67.38 | 71.88 | 75.56 | 77.4 |
#### ImageNet Linear Evaluation
The **Feature1 - Feature5** don't have the GlobalAveragePooling, the feature map is pooled to the specific dimensions and then follows a Linear layer to do the classification. Please refer to [resnet50_mhead_8xb32-steplr-90e.py](../../benchmarks/classification/imagenet/resnet50_mhead_8xb32-steplr-90e_in1k.py) for details of config.
The **Feature1 - Feature5** don't have the GlobalAveragePooling, the feature map is pooled to the specific dimensions and then follows a Linear layer to do the classification. Please refer to [resnet50_mhead_8xb32-steplr-90e.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/classification/imagenet/resnet50_mhead_8xb32-steplr-90e_in1k.py) for details of config.
The **AvgPool** result is obtained from Linear Evaluation with GlobalAveragePooling. Please refer to [resnet50_8xb32-steplr-100e_in1k](../../benchmarks/classification/imagenet/resnet50_8xb32-steplr-100e_in1k.py) for details of config.
The **AvgPool** result is obtained from Linear Evaluation with GlobalAveragePooling. Please refer to [resnet50_8xb32-steplr-100e_in1k.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/classification/imagenet/resnet50_8xb32-steplr-100e_in1k.py) for details of config.
| Self-Supervised Config | Feature1 | Feature2 | Feature3 | Feature4 | Feature5 | AvgPool |
| --------------------------------------------------------------------- | -------- | -------- | -------- | -------- | -------- | ------- |
| [resnet50_8xb32-coslr-200e](simclr_resnet50_8xb32-coslr-200e_in1k.py) | 14.43 | 30.97 | 41.02 | 53.92 | 61.24 | 58.92 |
| Self-Supervised Config | Feature1 | Feature2 | Feature3 | Feature4 | Feature5 | AvgPool |
| ------------------------------------------------------------------------------------------------------------------------------------------------ | -------- | -------- | -------- | -------- | -------- | ------- |
| [resnet50_8xb32-coslr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simclr/simclr_resnet50_8xb32-coslr-200e_in1k.py) | 14.43 | 30.97 | 41.02 | 53.92 | 61.24 | 57.28 |
#### Places205 Linear Evaluation
The **Feature1 - Feature5** don't have the GlobalAveragePooling, the feature map is pooled to the specific dimensions and then follows a Linear layer to do the classification. Please refer to [resnet50_mhead_8xb32-steplr-28e_places205.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/classification/places205/resnet50_mhead_8xb32-steplr-28e_places205.py) for details of config.
| Self-Supervised Config | Feature1 | Feature2 | Feature3 | Feature4 | Feature5 |
| ------------------------------------------------------------------------------------------------------------------------------------------------ | -------- | -------- | -------- | -------- | -------- |
| [resnet50_8xb32-coslr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simclr/simclr_resnet50_8xb32-coslr-200e_in1k.py) | 20.60 | 33.62 | 38.86 | 45.25 | 50.91 |
### Detection
@ -48,19 +56,19 @@ The detection benchmarks includes 2 downstream task datasets, **Pascal VOC 2007
#### Pascal VOC 2007 + 2012
Please refer to [faster_rcnn_r50_c4_mstrain_24k_voc0712.py](../../benchmarks/mmdetection/voc0712/faster_rcnn_r50_c4_mstrain_24k_voc0712.py) for details of config.
Please refer to [faster_rcnn_r50_c4_mstrain_24k_voc0712.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/mmdetection/voc0712/faster_rcnn_r50_c4_mstrain_24k_voc0712.py) for details of config.
| Self-Supervised Config | AP50 |
| --------------------------------------------------------------------- | ----- |
| [resnet50_8xb32-coslr-200e](simclr_resnet50_8xb32-coslr-200e_in1k.py) | 79.38 |
| Self-Supervised Config | AP50 |
| ------------------------------------------------------------------------------------------------------------------------------------------------ | ----- |
| [resnet50_8xb32-coslr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simclr/simclr_resnet50_8xb32-coslr-200e_in1k.py) | 79.38 |
#### COCO2017
Please refer to [mask_rcnn_r50_fpn_mstrain_1x_coco.py](../../benchmarks/mmdetection/coco/mask_rcnn_r50_fpn_mstrain_1x_coco.py) for details of config.
Please refer to [mask_rcnn_r50_fpn_mstrain_1x_coco.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/mmdetection/coco/mask_rcnn_r50_fpn_mstrain_1x_coco.py) for details of config.
| Self-Supervised Config | mAP(Box) | AP50(Box) | AP75(Box) | mAP(Mask) | AP50(Mask) | AP75(Mask) |
| --------------------------------------------------------------------- | -------- | --------- | --------- | --------- | ---------- | ---------- |
| [resnet50_8xb32-coslr-200e](simclr_resnet50_8xb32-coslr-200e_in1k.py) | 38.7 | 58.1 | 42.4 | 34.9 | 55.3 | 37.5 |
| Self-Supervised Config | mAP(Box) | AP50(Box) | AP75(Box) | mAP(Mask) | AP50(Mask) | AP75(Mask) |
| ------------------------------------------------------------------------------------------------------------------------------------------------ | -------- | --------- | --------- | --------- | ---------- | ---------- |
| [resnet50_8xb32-coslr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simclr/simclr_resnet50_8xb32-coslr-200e_in1k.py) | 38.7 | 58.1 | 42.4 | 34.9 | 55.3 | 37.5 |
### Segmentation
@ -68,11 +76,11 @@ The segmentation benchmarks includes 2 downstream task datasets, **Cityscapes**
#### Pascal VOC 2012 + Aug
Please refer to [fcn_r50-d8_512x512_20k_voc12aug.py](../../benchmarks/mmsegmentation/voc12aug/fcn_r50-d8_512x512_20k_voc12aug.py) for details of config.
Please refer to [fcn_r50-d8_512x512_20k_voc12aug.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/mmsegmentation/voc12aug/fcn_r50-d8_512x512_20k_voc12aug.py) for details of config.
| Self-Supervised Config | mIOU |
| --------------------------------------------------------------------- | ----- |
| [resnet50_8xb32-coslr-200e](simclr_resnet50_8xb32-coslr-200e_in1k.py) | 64.03 |
| Self-Supervised Config | mIOU |
| ------------------------------------------------------------------------------------------------------------------------------------------------ | ----- |
| [resnet50_8xb32-coslr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simclr/simclr_resnet50_8xb32-coslr-200e_in1k.py) | 64.03 |
## Citation

4
configs/selfsup/simclr/metafile.yml
View File

@ -23,6 +23,6 @@ Models:
- Task: Self-Supervised Image Classification
Dataset: ImageNet-1k
Metrics:
Top 1 Accuracy: 58.92
Top 1 Accuracy: 57.28
Config: configs/selfsup/simclr/simclr_resnet50_8xb32-coslr-200e_in1k.py
Weights: https://download.openmmlab.com/mmselfsup/simclr/simclr_resnet50_8xb32-coslr-200e_in1k_20211213-d0e53669.pth
Weights: https://download.openmmlab.com/mmselfsup/simclr/simclr_resnet50_8xb32-coslr-200e_in1k_20220225-97d2abef.pth

60
configs/selfsup/simsiam/README.md
View File

@ -14,7 +14,7 @@ Siamese networks have become a common structure in various recent models for uns
## Results and Models
**Back to [model_zoo.md](../../../docs/en/model_zoo.md) to download models.**
**Back to [model_zoo.md](https://github.com/open-mmlab/mmselfsup/blob/master/docs/en/model_zoo.md) to download models.**
In this page, we provide benchmarks as much as possible to evaluate our pre-trained models. If not mentioned, all models were trained on ImageNet1k dataset.
@ -28,22 +28,30 @@ The **Best Layer** indicates that the best results are obtained from which layer
Besides, k=1 to 96 indicates the hyper-parameter of Low-shot SVM.
| Self-Supervised Config | Best Layer | SVM | k=1 | k=2 | k=4 | k=8 | k=16 | k=32 | k=64 | k=96 |
| ---------------------------------------------------------------------- | ---------- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- |
| [resnet50_8xb32-coslr-100e](simsiam_resnet50_8xb32-coslr-100e_in1k.py) | feature5 | 84.64 | 39.65 | 49.86 | 62.48 | 69.50 | 74.48 | 78.31 | 81.06 | 82.56 |
| [resnet50_8xb32-coslr-200e](simsiam_resnet50_8xb32-coslr-200e_in1k.py) | feature5 | 85.20 | 39.85 | 50.44 | 63.73 | 70.93 | 75.74 | 79.42 | 82.02 | 83.44 |
| Self-Supervised Config | Best Layer | SVM | k=1 | k=2 | k=4 | k=8 | k=16 | k=32 | k=64 | k=96 |
| -------------------------------------------------------------------------------------------------------------------------------------------------- | ---------- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- |
| [resnet50_8xb32-coslr-100e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-100e_in1k.py) | feature5 | 84.64 | 39.65 | 49.86 | 62.48 | 69.50 | 74.48 | 78.31 | 81.06 | 82.56 |
| [resnet50_8xb32-coslr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-200e_in1k.py) | feature5 | 85.20 | 39.85 | 50.44 | 63.73 | 70.93 | 75.74 | 79.42 | 82.02 | 83.44 |
#### ImageNet Linear Evaluation
The **Feature1 - Feature5** don't have the GlobalAveragePooling, the feature map is pooled to the specific dimensions and then follows a Linear layer to do the classification. Please refer to [resnet50_mhead_8xb32-steplr-90e.py](../../benchmarks/classification/imagenet/resnet50_mhead_8xb32-steplr-90e_in1k.py) for details of config.
The **Feature1 - Feature5** don't have the GlobalAveragePooling, the feature map is pooled to the specific dimensions and then follows a Linear layer to do the classification. Please refer to [resnet50_mhead_8xb32-steplr-90e.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/classification/imagenet/resnet50_mhead_8xb32-steplr-90e_in1k.py) for details of config.
The **AvgPool** result is obtained from Linear Evaluation with GlobalAveragePooling. Please refer to [resnet50_8xb512-coslr-90e_in1k](../../benchmarks/classification/imagenet/resnet50_8xb512-coslr-90e_in1k.py) for details of config.
The **AvgPool** result is obtained from Linear Evaluation with GlobalAveragePooling. Please refer to [resnet50_8xb512-coslr-90e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/classification/imagenet/resnet50_8xb512-coslr-90e_in1k.py) for details of config.
| Self-Supervised Config | Feature1 | Feature2 | Feature3 | Feature4 | Feature5 | AvgPool |
| ---------------------------------------------------------------------- | -------- | -------- | -------- | -------- | -------- | ------- |
| [resnet50_8xb32-coslr-100e](simsiam_resnet50_8xb32-coslr-100e_in1k.py) | 16.27 | 33.77 | 45.80 | 60.83 | 68.21 | 68.20 |
| [resnet50_8xb32-coslr-200e](simsiam_resnet50_8xb32-coslr-200e_in1k.py) | 15.57 | 37.21 | 47.28 | 62.21 | 69.85 | 69.80 |
| Self-Supervised Config | Feature1 | Feature2 | Feature3 | Feature4 | Feature5 | AvgPool |
| -------------------------------------------------------------------------------------------------------------------------------------------------- | -------- | -------- | -------- | -------- | -------- | ------- |
| [resnet50_8xb32-coslr-100e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-100e_in1k.py) | 16.27 | 33.77 | 45.80 | 60.83 | 68.21 | 68.28 |
| [resnet50_8xb32-coslr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-200e_in1k.py) | 15.57 | 37.21 | 47.28 | 62.21 | 69.85 | 69.84 |
#### Places205 Linear Evaluation
The **Feature1 - Feature5** don't have the GlobalAveragePooling, the feature map is pooled to the specific dimensions and then follows a Linear layer to do the classification. Please refer to [resnet50_mhead_8xb32-steplr-28e_places205.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/classification/places205/resnet50_mhead_8xb32-steplr-28e_places205.py) for details of config.
| Self-Supervised Config | Feature1 | Feature2 | Feature3 | Feature4 | Feature5 |
| -------------------------------------------------------------------------------------------------------------------------------------------------- | -------- | -------- | -------- | -------- | -------- |
| [resnet50_8xb32-coslr-100e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-100e_in1k.py) | 21.32 | 35.66 | 43.05 | 50.79 | 53.27 |
| [resnet50_8xb32-coslr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-200e_in1k.py) | 21.17 | 35.85 | 43.49 | 50.99 | 54.10 |
### Detection
@ -51,21 +59,21 @@ The detection benchmarks includes 2 downstream task datasets, **Pascal VOC 2007
#### Pascal VOC 2007 + 2012
Please refer to [faster_rcnn_r50_c4_mstrain_24k_voc0712.py](../../benchmarks/mmdetection/voc0712/faster_rcnn_r50_c4_mstrain_24k_voc0712.py) for details of config.
Please refer to [faster_rcnn_r50_c4_mstrain_24k_voc0712.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/mmdetection/voc0712/faster_rcnn_r50_c4_mstrain_24k_voc0712.py) for details of config.
| Self-Supervised Config | AP50 |
| ---------------------------------------------------------------------- | ----- |
| [resnet50_8xb32-coslr-100e](simsiam_resnet50_8xb32-coslr-100e_in1k.py) | 79.80 |
| [resnet50_8xb32-coslr-200e](simsiam_resnet50_8xb32-coslr-200e_in1k.py) | 79.85 |
| Self-Supervised Config | AP50 |
| -------------------------------------------------------------------------------------------------------------------------------------------------- | ----- |
| [resnet50_8xb32-coslr-100e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-100e_in1k.py) | 79.80 |
| [resnet50_8xb32-coslr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-200e_in1k.py) | 79.85 |
#### COCO2017
Please refer to [mask_rcnn_r50_fpn_mstrain_1x_coco.py](../../benchmarks/mmdetection/coco/mask_rcnn_r50_fpn_mstrain_1x_coco.py) for details of config.
Please refer to [mask_rcnn_r50_fpn_mstrain_1x_coco.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/mmdetection/coco/mask_rcnn_r50_fpn_mstrain_1x_coco.py) for details of config.
| Self-Supervised Config | mAP(Box) | AP50(Box) | AP75(Box) | mAP(Mask) | AP50(Mask) | AP75(Mask) |
| ---------------------------------------------------------------------- | -------- | --------- | --------- | --------- | ---------- | ---------- |
| [resnet50_8xb32-coslr-100e](simsiam_resnet50_8xb32-coslr-100e_in1k.py) | 38.6 | 57.6 | 42.3 | 34.6 | 54.8 | 36.9 |
| [resnet50_8xb32-coslr-200e](simsiam_resnet50_8xb32-coslr-200e_in1k.py) | 38.8 | 58.0 | 42.3 | 34.9 | 55.3 | 37.6 |
| Self-Supervised Config | mAP(Box) | AP50(Box) | AP75(Box) | mAP(Mask) | AP50(Mask) | AP75(Mask) |
| -------------------------------------------------------------------------------------------------------------------------------------------------- | -------- | --------- | --------- | --------- | ---------- | ---------- |
| [resnet50_8xb32-coslr-100e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-100e_in1k.py) | 38.6 | 57.6 | 42.3 | 34.6 | 54.8 | 36.9 |
| [resnet50_8xb32-coslr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-200e_in1k.py) | 38.8 | 58.0 | 42.3 | 34.9 | 55.3 | 37.6 |
### Segmentation
@ -73,12 +81,12 @@ The segmentation benchmarks includes 2 downstream task datasets, **Cityscapes**
#### Pascal VOC 2012 + Aug
Please refer to [fcn_r50-d8_512x512_20k_voc12aug.py](../../benchmarks/mmsegmentation/voc12aug/fcn_r50-d8_512x512_20k_voc12aug.py) for details of config.
Please refer to [fcn_r50-d8_512x512_20k_voc12aug.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/mmsegmentation/voc12aug/fcn_r50-d8_512x512_20k_voc12aug.py) for details of config.
| Self-Supervised Config | mIOU |
| ---------------------------------------------------------------------- | ----- |
| [resnet50_8xb32-coslr-100e](simsiam_resnet50_8xb32-coslr-100e_in1k.py) | 48.35 |
| [resnet50_8xb32-coslr-200e](simsiam_resnet50_8xb32-coslr-200e_in1k.py) | 46.27 |
| Self-Supervised Config | mIOU |
| -------------------------------------------------------------------------------------------------------------------------------------------------- | ----- |
| [resnet50_8xb32-coslr-100e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-100e_in1k.py) | 48.35 |
| [resnet50_8xb32-coslr-200e](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-200e_in1k.py) | 46.27 |
## Citation

8
configs/selfsup/simsiam/metafile.yml
View File

@ -23,9 +23,9 @@ Models:
- Task: Self-Supervised Image Classification
Dataset: ImageNet-1k
Metrics:
Top 1 Accuracy: 68.20
Top 1 Accuracy: 68.28
Config: configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-100e_in1k.py
Weights: https://download.openmmlab.com/mmselfsup/simsiam/simsiam_resnet50_8xb32-coslr-100e_in1k_20211230-65a0eff4.pth
Weights: https://download.openmmlab.com/mmselfsup/simsiam/simsiam_resnet50_8xb32-coslr-100e_in1k_20220225-68a88ad8.pth
- Name: simsiam_resnet50_8xb32-coslr-200e_in1k
In Collection: SimSiam
Metadata:
@ -35,6 +35,6 @@ Models:
- Task: Self-Supervised Image Classification
Dataset: ImageNet-1k
Metrics:
Top 1 Accuracy: 69.80
Top 1 Accuracy: 69.84
Config: configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-200e_in1k.py
Weights: https://download.openmmlab.com/mmselfsup/simsiam/simsiam_resnet50_8xb32-coslr-200e_in1k_20211213-b605f9f1.pth
Weights: https://download.openmmlab.com/mmselfsup/simsiam/simsiam_resnet50_8xb32-coslr-200e_in1k_20220225-2f488143.pth

50
configs/selfsup/swav/README.md
View File

@ -14,7 +14,7 @@ Unsupervised image representations have significantly reduced the gap with super
## Results and Models
**Back to [model_zoo.md](../../../docs/en/model_zoo.md) to download models.**
**Back to [model_zoo.md](https://github.com/open-mmlab/mmselfsup/blob/master/docs/en/model_zoo.md) to download models.**
In this page, we provide benchmarks as much as possible to evaluate our pre-trained models. If not mentioned, all models were trained on ImageNet1k dataset.
@ -28,19 +28,27 @@ The **Best Layer** indicates that the best results are obtained from which layer
Besides, k=1 to 96 indicates the hyper-parameter of Low-shot SVM.
| Self-Supervised Config | Best Layer | SVM | k=1 | k=2 | k=4 | k=8 | k=16 | k=32 | k=64 | k=96 |
| ---------------------------------------------------------------------------------------------------------- | ---------- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- |
| [resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96](swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96.py) | feature5 | 87.00 | 44.68 | 55.41 | 67.64 | 73.67 | 78.14 | 81.58 | 83.98 | 85.15 |
| Self-Supervised Config | Best Layer | SVM | k=1 | k=2 | k=4 | k=8 | k=16 | k=32 | k=64 | k=96 |
| ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ---------- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- |
| [resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/swav/swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96.py) | feature5 | 87.00 | 44.68 | 55.41 | 67.64 | 73.67 | 78.14 | 81.58 | 83.98 | 85.15 |
#### ImageNet Linear Evaluation
The **Feature1 - Feature5** don't have the GlobalAveragePooling, the feature map is pooled to the specific dimensions and then follows a Linear layer to do the classification. Please refer to [resnet50_mhead_8xb32-steplr-90e.py](../../benchmarks/classification/imagenet/resnet50_mhead_8xb32-steplr-90e_in1k.py) for details of config.
The **Feature1 - Feature5** don't have the GlobalAveragePooling, the feature map is pooled to the specific dimensions and then follows a Linear layer to do the classification. Please refer to [resnet50_mhead_8xb32-steplr-90e.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/classification/imagenet/resnet50_mhead_8xb32-steplr-90e_in1k.py) for details of config.
The **AvgPool** result is obtained from Linear Evaluation with GlobalAveragePooling. Please refer to [resnet50_8xb32-coslr-100e_in1k](../../benchmarks/classification/imagenet/resnet50_8xb32-coslr-100e_in1k.py) for details of config.
The **AvgPool** result is obtained from Linear Evaluation with GlobalAveragePooling. Please refer to [resnet50_8xb32-coslr-100e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/classification/imagenet/resnet50_8xb32-coslr-100e_in1k.py) for details of config.
| Self-Supervised Config | Feature1 | Feature2 | Feature3 | Feature4 | Feature5 | AvgPool |
| ---------------------------------------------------------------------------------------------------------- | -------- | -------- | -------- | -------- | -------- | ------- |
| [resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96](swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96.py) | 16.98 | 34.96 | 49.26 | 65.98 | 70.74 | 70.55 |
| Self-Supervised Config | Feature1 | Feature2 | Feature3 | Feature4 | Feature5 | AvgPool |
| ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | -------- | -------- | -------- | -------- | -------- | ------- |
| [resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/swav/swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96.py) | 16.98 | 34.96 | 49.26 | 65.98 | 70.74 | 70.47 |
#### Places205 Linear Evaluation
The **Feature1 - Feature5** don't have the GlobalAveragePooling, the feature map is pooled to the specific dimensions and then follows a Linear layer to do the classification. Please refer to [resnet50_mhead_8xb32-steplr-28e_places205.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/classification/places205/resnet50_mhead_8xb32-steplr-28e_places205.py) for details of config.
| Self-Supervised Config | Feature1 | Feature2 | Feature3 | Feature4 | Feature5 |
| ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | -------- | -------- | -------- | -------- | -------- |
| [resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/swav/swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96.py) | 23.33 | 35.45 | 43.13 | 51.98 | 55.09 |
### Detection
@ -48,19 +56,19 @@ The detection benchmarks includes 2 downstream task datasets, **Pascal VOC 2007
#### Pascal VOC 2007 + 2012
Please refer to [faster_rcnn_r50_c4_mstrain_24k_voc0712.py](../../benchmarks/mmdetection/voc0712/faster_rcnn_r50_c4_mstrain_24k_voc0712.py) for details of config.
Please refer to [faster_rcnn_r50_c4_mstrain_24k_voc0712.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/mmdetection/voc0712/faster_rcnn_r50_c4_mstrain_24k_voc0712.py) for details of config.
| Self-Supervised Config | AP50 |
| ---------------------------------------------------------------------------------------------------------- | ----- |
| [resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96](swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96.py) | 77.64 |
| Self-Supervised Config | AP50 |
| ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----- |
| [resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/swav/swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96.py) | 77.64 |
#### COCO2017
Please refer to [mask_rcnn_r50_fpn_mstrain_1x_coco.py](../../benchmarks/mmdetection/coco/mask_rcnn_r50_fpn_mstrain_1x_coco.py) for details of config.
Please refer to [mask_rcnn_r50_fpn_mstrain_1x_coco.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/mmdetection/coco/mask_rcnn_r50_fpn_mstrain_1x_coco.py) for details of config.
| Self-Supervised Config | mAP(Box) | AP50(Box) | AP75(Box) | mAP(Mask) | AP50(Mask) | AP75(Mask) |
| ---------------------------------------------------------------------------------------------------------- | -------- | --------- | --------- | --------- | ---------- | ---------- |
| [resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96](swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96.py) | 40.2 | 60.5 | 43.9 | 36.3 | 57.5 | 38.8 |
| Self-Supervised Config | mAP(Box) | AP50(Box) | AP75(Box) | mAP(Mask) | AP50(Mask) | AP75(Mask) |
| ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | -------- | --------- | --------- | --------- | ---------- | ---------- |
| [resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/swav/swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96.py) | 40.2 | 60.5 | 43.9 | 36.3 | 57.5 | 38.8 |
### Segmentation
@ -68,11 +76,11 @@ The segmentation benchmarks includes 2 downstream task datasets, **Cityscapes**
#### Pascal VOC 2012 + Aug
Please refer to [fcn_r50-d8_512x512_20k_voc12aug.py](../../benchmarks/mmsegmentation/voc12aug/fcn_r50-d8_512x512_20k_voc12aug.py) for details of config.
Please refer to [fcn_r50-d8_512x512_20k_voc12aug.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/mmsegmentation/voc12aug/fcn_r50-d8_512x512_20k_voc12aug.py) for details of config.
| Self-Supervised Config | mIOU |
| ---------------------------------------------------------------------------------------------------------- | ----- |
| [resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96](swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96.py) | 63.73 |
| Self-Supervised Config | mIOU |
| ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----- |
| [resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/swav/swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96.py) | 63.73 |
## Citation

4
configs/selfsup/swav/metafile.yml
View File

@ -23,6 +23,6 @@ Models:
- Task: Self-Supervised Image Classification
Dataset: ImageNet-1k
Metrics:
Top 1 Accuracy: 70.55
Top 1 Accuracy: 70.47
Config: configs/selfsup/swav/swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96.py
Weights: https://download.openmmlab.com/mmselfsup/swav/swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96_20211213-0028900c.pth
Weights: https://download.openmmlab.com/mmselfsup/swav/swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96_20220225-0497dd5d.pth

27
docs/en/changelog.md
View File

@ -2,6 +2,33 @@
## MMSelfSup
### v0.7.0 (03/03/2022)
#### Highlight
* Support MAE ([#221](https://github.com/open-mmlab/mmselfsup/pull/221))
* Add Places205 benchmarks ([#210](https://github.com/open-mmlab/mmselfsup/pull/210))
* Add test Windows in workflows ([#215](https://github.com/open-mmlab/mmselfsup/pull/215))
#### New Features
* Support MAE ([#221](https://github.com/open-mmlab/mmselfsup/pull/221))
* Add Places205 benchmarks ([#210](https://github.com/open-mmlab/mmselfsup/pull/210))
#### Bug Fixes
* Fix config typos for rotation prediction and deepcluster ([#200](https://github.com/open-mmlab/mmselfsup/pull/200))
* Fix image channel bgr/rgb bug and update benchmarks ([#210](https://github.com/open-mmlab/mmselfsup/pull/210))
* Fix the bug when using prefetch under multi-view methods ([#218](https://github.com/open-mmlab/mmselfsup/pull/218))
* Fix tsne 'no init_cfg' error ([#222](https://github.com/open-mmlab/mmselfsup/pull/222))
#### Improvements
* Deprecate `imgs_per_gpu` and use `samples_per_gpu` ([#204](https://github.com/open-mmlab/mmselfsup/pull/204))
* Update the installation of MMCV ([#208](https://github.com/open-mmlab/mmselfsup/pull/208))
* Add pre-commit hook for algo-readme and copyright ([#213](https://github.com/open-mmlab/mmselfsup/pull/213))
* Add test Windows in workflows ([#215](https://github.com/open-mmlab/mmselfsup/pull/215))
#### Docs
* Translate 0_config.md into Chinese ([#216](https://github.com/open-mmlab/mmselfsup/pull/216))
* Reorganizing OpenMMLab projects and update algorithms in readme ([#219](https://github.com/open-mmlab/mmselfsup/pull/219))
### v0.6.0 (02/02/2022)
#### Highlight

3
docs/en/install.md
View File

@ -16,7 +16,8 @@ Compatible MMCV, MMClassification, MMDetection and MMSegmentation versions are s
| MMSelfSup version | MMCV version | MMClassification version | MMSegmentation version | MMDetection version |
| :---------------: | :-----------------: | :----------------------: | :--------------------: | :-----------------: |
| 0.6.0 (master) | mmcv-full >= 1.3.16 | mmcls >= 0.19.0 | mmseg >= 0.20.2 | mmdet >= 2.16.0 |
| 0.7.0 (master) | mmcv-full >= 1.3.16 | mmcls >= 0.19.0 | mmseg >= 0.20.2 | mmdet >= 2.16.0 |
| 0.6.0 | mmcv-full >= 1.3.16 | mmcls >= 0.19.0 | mmseg >= 0.20.2 | mmdet >= 2.16.0 |
| 0.5.0 | mmcv-full >= 1.3.16 | / | mmseg >= 0.20.2 | mmdet >= 2.16.0 |
**Note:**

82
docs/en/model_zoo.md
View File

@ -6,20 +6,21 @@ All models and part of benchmark results are recorded below.
| Algorithm | Config | Download |
| ------------------------------------------------------------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| [BYOL](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/byol/README.md) | [byol_resnet50_8xb32-accum16-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/byol/byol_resnet50_8xb32-accum16-coslr-200e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/byol/byol_resnet50_8xb32-accum16-coslr-200e_in1k_20211213-30dbaef1.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/byol/byol_resnet50_8xb32-accum16-coslr-200e_in1k_20211111_212813.log.json) |
| | [byol_resnet50_8xb32-accum16-coslr-300e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/byol/byol_resnet50_8xb32-accum16-coslr-300e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/byol/byol_resnet50_8xb32-accum16-coslr-300e_in1k_20211213-47673e22.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/byol/byol_resnet50_8xb32-accum16-coslr-300e_in1k_20211129_163841.log.json) |
| [BYOL](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/byol/README.md) | [byol_resnet50_8xb32-accum16-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/byol/byol_resnet50_8xb32-accum16-coslr-200e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/byol/byol_resnet50_8xb32-accum16-coslr-200e_in1k_20220225-5c8b2c2e.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/byol/byol_resnet50_8xb32-accum16-coslr-200e_in1k_20220214_115709.log.json) |
| | [byol_resnet50_8xb32-accum16-coslr-300e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/byol/byol_resnet50_8xb32-accum16-coslr-300e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/byol/byol_resnet50_8xb32-accum16-coslr-300e_in1k_20220225-a0daa54a.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/byol/byol_resnet50_8xb32-accum16-coslr-300e_in1k_20220210_095852.log.json) |
| [DeepCluster](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/deepcluster/README.md) | [deepcluster-sobel_resnet50_8xb64-steplr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/deepcluster/deepcluster-sobel_resnet50_8xb64-steplr-200e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/deepcluster/deepcluster-sobel_resnet50_8xb64-steplr-200e_in1k-bb8681e2.pth) |
| [DenseCL](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/densecl/README.md) | [densecl_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/densecl/densecl_resnet50_8xb32-coslr-200e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/densecl/densecl_resnet50_8xb32-coslr-200e_in1k_20211214-1efb342c.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/densecl/densecl_resnet50_8xb32-coslr-200e_in1k_20211210_230413.log.json) |
| [MoCo v2](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mocov2/README.md) | [mocov2_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mocov2/mocov2_resnet50_8xb32-coslr-200e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/moco/mocov2_resnet50_8xb32-coslr-200e_in1k_20211213-7ce8f840.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/moco/mocov2_resnet50_8xb32-coslr-200e_in1k_20211208_161634.log.json) |
| [NPID](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/npid/README.md) | [npid_resnet50_8xb32-steplr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/npid/npid_resnet50_8xb32-steplr-200e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/npid/npid_resnet50_8xb32-steplr-200e_in1k_20211213-b5fec6df.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/npid/npid_resnet50_8xb32-steplr-200e_in1k_20211210_124652.log.json) |
| [ODC](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/odc/README.md) | [odc_resnet50_8xb64-steplr-440e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/odc/odc_resnet50_8xb64-steplr-440e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/odc/odc_resnet50_8xb64-steplr-440e_in1k-5af5dd0c.pth) |
| [Relative Location](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/relative_loc/README.md) | [relative-loc_resnet50_8xb64-steplr-70e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/relative_loc/relative-loc_resnet50_8xb64-steplr-70e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/relative_loc/relative-loc_resnet50_8xb64-steplr-70e_in1k_20211213-cdd3162f.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/relative_loc/relative-loc_resnet50_8xb64-steplr-70e_in1k_20210930_144754.log.json) |
| [Rotation Prediction](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/rotation_pred/README.md) | [rotation-pred_resnet50_8xb16-steplr-70e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/rotation_pred/rotation-pred_resnet50_8xb16-steplr-70e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/rotation_pred/rotation-pred_resnet50_8xb16-steplr-70e_in1k_20211213-513972ac.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/rotation_pred/rotation-pred_resnet50_8xb16-steplr-70e_in1k_20210930_151459.log.json) |
| [SimCLR](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simclr/README.md) | [simclr_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simclr/simclr_resnet50_8xb32-coslr-200e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/simclr/simclr_resnet50_8xb32-coslr-200e_in1k_20211213-d0e53669.pth) |
| [SimSiam](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/README.md) | [simsiam_resnet50_8xb32-coslr-100e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-100e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/simsiam/simsiam_resnet50_8xb32-coslr-100e_in1k_20211230-65a0eff4.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/simsiam/simsiam_resnet50_8xb32-coslr-100e_in1k_20211225_132004.log.json) |
| | [simsiam_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-200e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/simsiam/simsiam_resnet50_8xb32-coslr-200e_in1k_20211213-b605f9f1.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/simsiam/simsiam_resnet50_8xb32-coslr-200e_in1k_20211225_132031.log.json) |
| [SwAV](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/swav/README.md) | [swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/swav/swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96.py) | [model](https://download.openmmlab.com/mmselfsup/swav/swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96_20211213-0028900c.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/swav/swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96_20211206_102636.log.json) |
| [MoCo v3](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mocov3/README.md) | [mocov3_vit-small-p16_32xb128-fp16-coslr-300e_in1k-224](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mocov3/mocov3_vit-small-p16_32xb128-fp16-coslr-300e_in1k-224.py) | [model](https://download.openmmlab.com/mmselfsup/moco/mocov3_vit-small-p16_32xb128-fp16-coslr-300e_in1k-224_20220127-e9332db2.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/moco/mocov3_vit-small-p16_32xb128-fp16-coslr-300e_in1k-224_20220115_171742.log.json) |
| [DenseCL](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/densecl/README.md) | [densecl_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/densecl/densecl_resnet50_8xb32-coslr-200e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/densecl/densecl_resnet50_8xb32-coslr-200e_in1k_20220225-8c7808fe.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/densecl/densecl_resnet50_8xb32-coslr-200e_in1k_20220215_041207.log.json) |
| [MoCo v2](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mocov2/README.md) | [mocov2_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mocov2/mocov2_resnet50_8xb32-coslr-200e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/moco/mocov2_resnet50_8xb32-coslr-200e_in1k_20220225-89e03af4.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/moco/mocov2_resnet50_8xb32-coslr-200e_in1k_20220210_110905.log.json) |
| [NPID](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/npid/README.md) | [npid_resnet50_8xb32-steplr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/npid/npid_resnet50_8xb32-steplr-200e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/npid/npid_resnet50_8xb32-steplr-200e_in1k_20220225-5fbbda2a.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/npid/npid_resnet50_8xb32-steplr-200e_in1k_20220215_185513.log.json) |
| [ODC](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/odc/README.md) | [odc_resnet50_8xb64-steplr-440e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/odc/odc_resnet50_8xb64-steplr-440e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/odc/odc_resnet50_8xb64-steplr-440e_in1k_20220225-a755d9c0.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/odc/odc_resnet50_8xb64-steplr-440e_in1k_20220215_235245.log.json) |
| [Relative Location](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/relative_loc/README.md) | [relative-loc_resnet50_8xb64-steplr-70e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/relative_loc/relative-loc_resnet50_8xb64-steplr-70e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/relative_loc/relative-loc_resnet50_8xb64-steplr-70e_in1k_20220225-84784688.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/relative_loc/relative-loc_resnet50_8xb64-steplr-70e_in1k_20220211_124808.log.json) |
| [Rotation Prediction](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/rotation_pred/README.md) | [rotation-pred_resnet50_8xb16-steplr-70e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/rotation_pred/rotation-pred_resnet50_8xb16-steplr-70e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/rotation_pred/rotation-pred_resnet50_8xb16-steplr-70e_in1k_20220225-5b9f06a0.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/rotation_pred/rotation-pred_resnet50_8xb16-steplr-70e_in1k_20220215_185303.log.json) |
| [SimCLR](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simclr/README.md) | [simclr_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simclr/simclr_resnet50_8xb32-coslr-200e_in1k.py) | [model](simclr_resnet50_8xb32-coslr-200e_in1k_20220225-97d2abef.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/simclr/simclr_resnet50_8xb64-coslr-200e_in1k_20220210_191629.log.json) |
| [SimSiam](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/README.md) | [simsiam_resnet50_8xb32-coslr-100e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-100e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/simsiam/simsiam_resnet50_8xb32-coslr-100e_in1k_20220225-68a88ad8.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/simsiam/simsiam_resnet50_8xb32-coslr-100e_in1k_20220210_195405.log.json) |
| | [simsiam_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-200e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/simsiam/simsiam_resnet50_8xb32-coslr-200e_in1k_20220225-2f488143.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/simsiam/simsiam_resnet50_8xb32-coslr-200e_in1k_20220210_195402.log.json) |
| [SwAV](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/swav/README.md) | [swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/swav/swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96.py) | [model](https://download.openmmlab.com/mmselfsup/swav/swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96_20220225-0497dd5d.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/swav/swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96_20220211_061131.log.json) |
| [MoCo v3](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mocov3/README.md) | [mocov3_vit-small-p16_32xb128-fp16-coslr-300e_in1k-224](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mocov3/mocov3_vit-small-p16_32xb128-fp16-coslr-300e_in1k-224.py) | [model](https://download.openmmlab.com/mmselfsup/moco/mocov3_vit-small-p16_32xb128-fp16-coslr-300e_in1k-224_20220225-e31238dd.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/moco/mocov3_vit-small-p16_32xb128-fp16-coslr-300e_in1k-224_20220222_160222.log.json) |
| [MAE](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mae/README.md) | [mae_vit-base-p16_8xb512-coslr-400e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mae/mae_vit-base-p16_8xb512-coslr-400e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/mae/mae_vit-base-p16_8xb512-coslr-400e_in1k-224_20220223-85be947b.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/mae/mae_vit-base-p16_8xb512-coslr-300e_in1k-224_20220210_140925.log.json) |
Remarks:
@ -37,19 +38,26 @@ If not specified, we use linear evaluation setting from [MoCo](http://openaccess
| Algorithm | Config | Remarks | Top-1 (%) |
| ------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | --------------------- | --------- |
| BYOL | [byol_resnet50_8xb32-accum16-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/byol/byol_resnet50_8xb32-accum16-coslr-200e_in1k.py) | | 67.68 |
| BYOL | [byol_resnet50_8xb32-accum16-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/byol/byol_resnet50_8xb32-accum16-coslr-200e_in1k.py) | | 67.55 |
| | [byol_resnet50_8xb32-accum16-coslr-300e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/byol/byol_resnet50_8xb32-accum16-coslr-300e_in1k.py) | | 68.55 |
| DeepCluster | [deepcluster-sobel_resnet50_8xb64-steplr-200e_in1k.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/deepcluster/deepcluster-sobel_resnet50_8xb64-steplr-200e_in1k.py) | | 46.92 |
| DenseCL | [densecl_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/densecl/densecl_resnet50_8xb32-coslr-200e_in1k.py) | | 63.34 |
| MoCo v2 | [mocov2_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mocov2/mocov2_resnet50_8xb32-coslr-200e_in1k.py) | | 67.56 |
| NPID | [npid_resnet50_8xb32-steplr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/npid/npid_resnet50_8xb32-steplr-200e_in1k.py) | | 58.16 |
| ODC | [odc_resnet50_8xb64-steplr-440e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/odc/odc_resnet50_8xb64-steplr-440e_in1k.py) | | 53.42 |
| Relative Location | [relative-loc_resnet50_8xb64-steplr-70e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/relative_loc/relative-loc_resnet50_8xb64-steplr-70e_in1k.py) | | 39.65 |
| Rotation Prediction | [rotation-pred_resnet50_8xb16-steplr-70e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/rotation_pred/rotation-pred_resnet50_8xb16-steplr-70e_in1k.py) | | 44.35 |
| SimCLR | [simclr_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simclr/simclr_resnet50_8xb32-coslr-200e_in1k.py) | | 58.92 |
| SimSiam | [simsiam_resnet50_8xb32-coslr-100e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-100e_in1k.py) | SimSiam paper setting | 68.20 |
| | [simsiam_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-200e_in1k.py) | SimSiam paper setting | 69.80 |
| SwAV | [swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/swav/swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96.py) | SwAV paper setting | 70.55 |
| MoCo v3 | [mocov3_vit-small-p16_32xb128-fp16-coslr-300e_in1k-224](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mocov3/mocov3_vit-small-p16_32xb128-fp16-coslr-300e_in1k-224.py) | MoCo v3 paper setting | 73.07 |
| DenseCL | [densecl_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/densecl/densecl_resnet50_8xb32-coslr-200e_in1k.py) | | 63.62 |
| MoCo v2 | [mocov2_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mocov2/mocov2_resnet50_8xb32-coslr-200e_in1k.py) | | 67.58 |
| NPID | [npid_resnet50_8xb32-steplr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/npid/npid_resnet50_8xb32-steplr-200e_in1k.py) | | 58.97 |
| ODC | [odc_resnet50_8xb64-steplr-440e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/odc/odc_resnet50_8xb64-steplr-440e_in1k.py) | | 53.43 |
| Relative Location | [relative-loc_resnet50_8xb64-steplr-70e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/relative_loc/relative-loc_resnet50_8xb64-steplr-70e_in1k.py) | | 38.78 |
| Rotation Prediction | [rotation-pred_resnet50_8xb16-steplr-70e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/rotation_pred/rotation-pred_resnet50_8xb16-steplr-70e_in1k.py) | | 48.12 |
| SimCLR | [simclr_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simclr/simclr_resnet50_8xb32-coslr-200e_in1k.py) | | 57.28 |
| SimSiam | [simsiam_resnet50_8xb32-coslr-100e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-100e_in1k.py) | SimSiam paper setting | 68.28 |
| | [simsiam_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-200e_in1k.py) | SimSiam paper setting | 69.84 |
| SwAV | [swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/swav/swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96.py) | SwAV paper setting | 70.47 |
| MoCo v3 | [mocov3_vit-small-p16_32xb128-fp16-coslr-300e_in1k-224](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mocov3/mocov3_vit-small-p16_32xb128-fp16-coslr-300e_in1k-224.py) | MoCo v3 paper setting | 73.19 |
### ImageNet Fine-tuning
| Algorithm | Config | Remarks | Top-1 (%) |
| --------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------- | --------- |
| MAE | [mae_vit-base-p16_8xb512-coslr-400e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mae/mae_vit-base-p16_8xb512-coslr-400e_in1k.py) | | 83.1 |
### COCO17 Object Detection
@ -72,15 +80,15 @@ In COCO17 Object detection task, we choose the evluation protocol from [MoCo](ht
In Pascal VOC12 Aug Segmentation task, we choose the evluation protocol from [MMSeg](https://github.com/open-mmlab/mmsegmentation), with FCN architecture, the results below are trained with the same [config](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/mmsegmentation/voc12aug/fcn_r50-d8_512x512_20k_voc12aug.py).
| Algorithm | Config | mIOU |
| ------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----- |
| BYOL | [byol_resnet50_8xb32-accum16-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/byol/byol_resnet50_8xb32-accum16-coslr-200e_in1k.py) | 67.16 |
| DenseCL | [densecl_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/densecl/densecl_resnet50_8xb32-coslr-200e_in1k.py) | 69.47 |
| MoCo v2 | [mocov2_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mocov2/mocov2_resnet50_8xb32-coslr-200e_in1k.py) | 67.55 |
| NPID | [npid_resnet50_8xb32-steplr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/npid/npid_resnet50_8xb32-steplr-200e_in1k.py) | 65.45 |
| Relative Location | [relative-loc_resnet50_8xb64-steplr-70e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/relative_loc/relative-loc_resnet50_8xb64-steplr-70e_in1k.py) | 63.49 |
| Rotation Prediction | [rotation-pred_resnet50_8xb16-steplr-70e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/rotation_pred/rotation-pred_resnet50_8xb16-steplr-70e_in1k.py) | 64.31 |
| SimCLR | [simclr_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simclr/simclr_resnet50_8xb32-coslr-200e_in1k.py) | 64.03 |
| SimSiam | [simsiam_resnet50_8xb32-coslr-100e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-100e_in1k.py) | 48.35 |
| | [simsiam_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-200e_in1k.py) | 46.27 |
| SwAV | [swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/swav/swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96.py) | 63.73 |
| Algorithm | Config | mIOU |
| ------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----- |
| BYOL | [byol_resnet50_8xb32-accum16-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/byol/byol_resnet50_8xb32-accum16-coslr-200e_in1k.py) | 67.16 |
| DenseCL | [densecl_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/densecl/densecl_resnet50_8xb32-coslr-200e_in1k.py) | 69.47 |
| MoCo v2 | [mocov2_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mocov2/mocov2_resnet50_8xb32-coslr-200e_in1k.py) | 67.55 |
| NPID | [npid_resnet50_8xb32-steplr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/npid/npid_resnet50_8xb32-steplr-200e_in1k.py) | 65.45 |
| Relative Location | [relative-loc_resnet50_8xb64-steplr-70e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/relative_loc/relative-loc_resnet50_8xb64-steplr-70e_in1k.py) | 63.49 |
| Rotation Prediction | [rotation-pred_resnet50_8xb16-steplr-70e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/rotation_pred/rotation-pred_resnet50_8xb16-steplr-70e_in1k.py) | 64.31 |
| SimCLR | [simclr_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simclr/simclr_resnet50_8xb32-coslr-200e_in1k.py) | 64.03 |
| SimSiam | [simsiam_resnet50_8xb32-coslr-100e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-100e_in1k.py) | 48.35 |
| | [simsiam_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-200e_in1k.py) | 46.27 |
| SwAV | [swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/swav/swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96.py) | 63.73 |

16
docs/en/tutorials/0_config.md
View File

@ -51,9 +51,8 @@ We follow the below convention to name config files. Contributors are advised to
### Module information
```
{backbone setting}_{neck setting}_{head_setting}
{backbone setting}-{neck setting}-{head_setting}
```
Here we use `'_'` to concatenate to make the name more readable.
The module information mainly includes the backbone information. E.g:
- `resnet50`
@ -61,7 +60,6 @@ The module information mainly includes the backbone information. E.g:
Or there are some special settings which is needed to be mentioned in the config name. E.g:
- `resnet50-nofrz`: In some downstream tasks，the backbone will not froze stages while training
- `simclr_resnet50_mocov2-neck`: In simclr, using mocov2-neck to train, not simclr its own neck
### Training information
@ -83,18 +81,6 @@ Data information contains the dataset, input size, etc. E.g:
### Config File Name Example
```
mocov2_resnet50_simclr-neck_8xb32-coslr-200e_in1k.py
```
- `mocov2`: Algorithm information
- `resnet50_simclr-neck`: Module information
- `resnet50`: Backbone
- `simclr-neck`: Using special neck，not the default mocov2-neck
- `8xb32-coslr-200e`: Training information
- `8xb32`: Use 8 GPUs in total，and the batch size is 32 per GPU
- `coslr`: Use cosine learning rate scheduler
- `200e`: Train the model for 200 epoch
- `in1k`: Data information, train on ImageNet1k dataset
```
swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96.py
```
- `swav`: Algorithm information

27
docs/zh_cn/changelog.md
View File

@ -2,6 +2,33 @@
## MMSelfSup
### v0.7.0 (03/03/2022)
#### 亮点
* 支持 MAE 算法 ([#221](https://github.com/open-mmlab/mmselfsup/pull/221))
* 增加 Places205 下游基准测试 ([#210](https://github.com/open-mmlab/mmselfsup/pull/210))
* 在 CI 工作流中添加 Windows 测试 ([#215](https://github.com/open-mmlab/mmselfsup/pull/215))
#### 新特性
* 支持 MAE 算法 ([#221](https://github.com/open-mmlab/mmselfsup/pull/221))
* 增加 Places205 下游基准测试 ([#210](https://github.com/open-mmlab/mmselfsup/pull/210))
#### Bug 修复
* 修复部分配置文件中的错误 ([#200](https://github.com/open-mmlab/mmselfsup/pull/200))
* 修复图像读取通道问题并更新相关结果 ([#210](https://github.com/open-mmlab/mmselfsup/pull/210))
* 修复在使用 prefetch 时，部分 dataset 输出格式不匹配的问题 ([#218](https://github.com/open-mmlab/mmselfsup/pull/218))
* 修复 t-sne 'no init_cfg' 的错误 ([#222](https://github.com/open-mmlab/mmselfsup/pull/222))
#### 改进
* 配置文件中弃用 `imgs_per_gpu`， 改用 `samples_per_gpu` ([#204](https://github.com/open-mmlab/mmselfsup/pull/204))
* 更新 MMCV 的安装方式 ([#208](https://github.com/open-mmlab/mmselfsup/pull/208))
* 为 算法 readme 和代码版权增加 pre-commit 钩子 ([#213](https://github.com/open-mmlab/mmselfsup/pull/213))
* 在 CI 工作流中添加 Windows 测试 ([#215](https://github.com/open-mmlab/mmselfsup/pull/215))
#### 文档
* 将 0_config.md 翻译成中文 ([#216](https://github.com/open-mmlab/mmselfsup/pull/216))
* 更新主页 OpenMMLab 项目和介绍 ([#219](https://github.com/open-mmlab/mmselfsup/pull/219))
### v0.6.0 (02/02/2022)
#### 亮点

3
docs/zh_cn/install.md
View File

@ -16,7 +16,8 @@
| MMSelfSup version | MMCV version | MMClassification version | MMSegmentation version | MMDetection version |
| :---------------: | :-----------------: | :----------------------: | :--------------------: | :-----------------: |
| 0.6.0 (master) | mmcv-full >= 1.3.16 | mmcls >= 0.19.0 | mmseg >= 0.20.2 | mmdet >= 2.16.0 |
| 0.7.0 (master) | mmcv-full >= 1.3.16 | mmcls >= 0.19.0 | mmseg >= 0.20.2 | mmdet >= 2.16.0 |
| 0.6.0 | mmcv-full >= 1.3.16 | mmcls >= 0.19.0 | mmseg >= 0.20.2 | mmdet >= 2.16.0 |
| 0.5.0 | mmcv-full >= 1.3.16 | / | mmseg >= 0.20.2 | mmdet >= 2.16.0 |

81
docs/zh_cn/model_zoo.md
View File

@ -6,20 +6,21 @@
| 算法 | 配置文件 | 下载链接 |
| ------------------------------------------------------------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| [BYOL](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/byol/README.md) | [byol_resnet50_8xb32-accum16-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/byol/byol_resnet50_8xb32-accum16-coslr-200e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/byol/byol_resnet50_8xb32-accum16-coslr-200e_in1k_20211213-30dbaef1.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/byol/byol_resnet50_8xb32-accum16-coslr-200e_in1k_20211111_212813.log.json) |
| | [byol_resnet50_8xb32-accum16-coslr-300e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/byol/byol_resnet50_8xb32-accum16-coslr-300e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/byol/byol_resnet50_8xb32-accum16-coslr-300e_in1k_20211213-47673e22.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/byol/byol_resnet50_8xb32-accum16-coslr-300e_in1k_20211129_163841.log.json) |
| [BYOL](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/byol/README.md) | [byol_resnet50_8xb32-accum16-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/byol/byol_resnet50_8xb32-accum16-coslr-200e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/byol/byol_resnet50_8xb32-accum16-coslr-200e_in1k_20220225-5c8b2c2e.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/byol/byol_resnet50_8xb32-accum16-coslr-200e_in1k_20220214_115709.log.json) |
| | [byol_resnet50_8xb32-accum16-coslr-300e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/byol/byol_resnet50_8xb32-accum16-coslr-300e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/byol/byol_resnet50_8xb32-accum16-coslr-300e_in1k_20220225-a0daa54a.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/byol/byol_resnet50_8xb32-accum16-coslr-300e_in1k_20220210_095852.log.json) |
| [DeepCluster](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/deepcluster/README.md) | [deepcluster-sobel_resnet50_8xb64-steplr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/deepcluster/deepcluster-sobel_resnet50_8xb64-steplr-200e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/deepcluster/deepcluster-sobel_resnet50_8xb64-steplr-200e_in1k-bb8681e2.pth) |
| [DenseCL](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/densecl/README.md) | [densecl_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/densecl/densecl_resnet50_8xb32-coslr-200e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/densecl/densecl_resnet50_8xb32-coslr-200e_in1k_20211214-1efb342c.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/densecl/densecl_resnet50_8xb32-coslr-200e_in1k_20211210_230413.log.json) |
| [MoCo v2](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mocov2/README.md) | [mocov2_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mocov2/mocov2_resnet50_8xb32-coslr-200e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/moco/mocov2_resnet50_8xb32-coslr-200e_in1k_20211213-7ce8f840.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/moco/mocov2_resnet50_8xb32-coslr-200e_in1k_20211208_161634.log.json) |
| [NPID](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/npid/README.md) | [npid_resnet50_8xb32-steplr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/npid/npid_resnet50_8xb32-steplr-200e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/npid/npid_resnet50_8xb32-steplr-200e_in1k_20211213-b5fec6df.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/npid/npid_resnet50_8xb32-steplr-200e_in1k_20211210_124652.log.json) |
| [ODC](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/odc/README.md) | [odc_resnet50_8xb64-steplr-440e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/odc/odc_resnet50_8xb64-steplr-440e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/odc/odc_resnet50_8xb64-steplr-440e_in1k-5af5dd0c.pth) |
| [Relative Location](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/relative_loc/README.md) | [relative-loc_resnet50_8xb64-steplr-70e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/relative_loc/relative-loc_resnet50_8xb64-steplr-70e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/relative_loc/relative-loc_resnet50_8xb64-steplr-70e_in1k_20211213-cdd3162f.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/relative_loc/relative-loc_resnet50_8xb64-steplr-70e_in1k_20210930_144754.log.json) |
| [Rotation Prediction](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/rotation_pred/README.md) | [rotation-pred_resnet50_8xb16-steplr-70e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/rotation_pred/rotation-pred_resnet50_8xb16-steplr-70e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/rotation_pred/rotation-pred_resnet50_8xb16-steplr-70e_in1k_20211213-513972ac.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/rotation_pred/rotation-pred_resnet50_8xb16-steplr-70e_in1k_20210930_151459.log.json) |
| [SimCLR](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simclr/README.md) | [simclr_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simclr/simclr_resnet50_8xb32-coslr-200e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/simclr/simclr_resnet50_8xb32-coslr-200e_in1k_20211213-d0e53669.pth) |
| [SimSiam](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/README.md) | [simsiam_resnet50_8xb32-coslr-100e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-100e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/simsiam/simsiam_resnet50_8xb32-coslr-100e_in1k_20211230-65a0eff4.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/simsiam/simsiam_resnet50_8xb32-coslr-100e_in1k_20211225_132004.log.json) |
| | [simsiam_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-200e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/simsiam/simsiam_resnet50_8xb32-coslr-200e_in1k_20211213-b605f9f1.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/simsiam/simsiam_resnet50_8xb32-coslr-200e_in1k_20211225_132031.log.json) |
| [SwAV](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/swav/README.md) | [swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/swav/swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96.py) | [model](https://download.openmmlab.com/mmselfsup/swav/swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96_20211213-0028900c.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/swav/swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96_20211206_102636.log.json) |
| [MoCo v3](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mocov3/README.md) | [mocov3_vit-small-p16_32xb128-fp16-coslr-300e_in1k-224](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mocov3/mocov3_vit-small-p16_32xb128-fp16-coslr-300e_in1k-224.py) | [model](https://download.openmmlab.com/mmselfsup/moco/mocov3_vit-small-p16_32xb128-fp16-coslr-300e_in1k-224_20220127-e9332db2.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/moco/mocov3_vit-small-p16_32xb128-fp16-coslr-300e_in1k-224_20220115_171742.log.json) |
| [DenseCL](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/densecl/README.md) | [densecl_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/densecl/densecl_resnet50_8xb32-coslr-200e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/densecl/densecl_resnet50_8xb32-coslr-200e_in1k_20220225-8c7808fe.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/densecl/densecl_resnet50_8xb32-coslr-200e_in1k_20220215_041207.log.json) |
| [MoCo v2](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mocov2/README.md) | [mocov2_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mocov2/mocov2_resnet50_8xb32-coslr-200e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/moco/mocov2_resnet50_8xb32-coslr-200e_in1k_20220225-89e03af4.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/moco/mocov2_resnet50_8xb32-coslr-200e_in1k_20220210_110905.log.json) |
| [NPID](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/npid/README.md) | [npid_resnet50_8xb32-steplr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/npid/npid_resnet50_8xb32-steplr-200e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/npid/npid_resnet50_8xb32-steplr-200e_in1k_20220225-5fbbda2a.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/npid/npid_resnet50_8xb32-steplr-200e_in1k_20220215_185513.log.json) |
| [ODC](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/odc/README.md) | [odc_resnet50_8xb64-steplr-440e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/odc/odc_resnet50_8xb64-steplr-440e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/odc/odc_resnet50_8xb64-steplr-440e_in1k_20220225-a755d9c0.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/odc/odc_resnet50_8xb64-steplr-440e_in1k_20220215_235245.log.json) |
| [Relative Location](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/relative_loc/README.md) | [relative-loc_resnet50_8xb64-steplr-70e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/relative_loc/relative-loc_resnet50_8xb64-steplr-70e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/relative_loc/relative-loc_resnet50_8xb64-steplr-70e_in1k_20220225-84784688.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/relative_loc/relative-loc_resnet50_8xb64-steplr-70e_in1k_20220211_124808.log.json) |
| [Rotation Prediction](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/rotation_pred/README.md) | [rotation-pred_resnet50_8xb16-steplr-70e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/rotation_pred/rotation-pred_resnet50_8xb16-steplr-70e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/rotation_pred/rotation-pred_resnet50_8xb16-steplr-70e_in1k_20220225-5b9f06a0.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/rotation_pred/rotation-pred_resnet50_8xb16-steplr-70e_in1k_20220215_185303.log.json) |
| [SimCLR](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simclr/README.md) | [simclr_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simclr/simclr_resnet50_8xb32-coslr-200e_in1k.py) | [model](simclr_resnet50_8xb32-coslr-200e_in1k_20220225-97d2abef.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/simclr/simclr_resnet50_8xb64-coslr-200e_in1k_20220210_191629.log.json) |
| [SimSiam](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/README.md) | [simsiam_resnet50_8xb32-coslr-100e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-100e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/simsiam/simsiam_resnet50_8xb32-coslr-100e_in1k_20220225-68a88ad8.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/simsiam/simsiam_resnet50_8xb32-coslr-100e_in1k_20220210_195405.log.json) |
| | [simsiam_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-200e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/simsiam/simsiam_resnet50_8xb32-coslr-200e_in1k_20220225-2f488143.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/simsiam/simsiam_resnet50_8xb32-coslr-200e_in1k_20220210_195402.log.json) |
| [SwAV](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/swav/README.md) | [swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/swav/swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96.py) | [model](https://download.openmmlab.com/mmselfsup/swav/swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96_20220225-0497dd5d.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/swav/swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96_20220211_061131.log.json) |
| [MoCo v3](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mocov3/README.md) | [mocov3_vit-small-p16_32xb128-fp16-coslr-300e_in1k-224](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mocov3/mocov3_vit-small-p16_32xb128-fp16-coslr-300e_in1k-224.py) | [model](https://download.openmmlab.com/mmselfsup/moco/mocov3_vit-small-p16_32xb128-fp16-coslr-300e_in1k-224_20220225-e31238dd.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/moco/mocov3_vit-small-p16_32xb128-fp16-coslr-300e_in1k-224_20220222_160222.log.json) |
| [MAE](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mae/README.md) | [mae_vit-base-p16_8xb512-coslr-400e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mae/mae_vit-base-p16_8xb512-coslr-400e_in1k.py) | [model](https://download.openmmlab.com/mmselfsup/mae/mae_vit-base-p16_8xb512-coslr-400e_in1k-224_20220223-85be947b.pth) &#124; [log](https://download.openmmlab.com/mmselfsup/mae/mae_vit-base-p16_8xb512-coslr-300e_in1k-224_20220210_140925.log.json) |
备注：
@ -37,20 +38,26 @@
| 算法 | 配置文件 | 备注 | Top-1 (%) |
| ------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ---------------- | --------- |
| BYOL | [byol_resnet50_8xb32-accum16-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/byol/byol_resnet50_8xb32-accum16-coslr-200e_in1k.py) | | 67.68 |
| BYOL | [byol_resnet50_8xb32-accum16-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/byol/byol_resnet50_8xb32-accum16-coslr-200e_in1k.py) | | 67.55 |
| | [byol_resnet50_8xb32-accum16-coslr-300e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/byol/byol_resnet50_8xb32-accum16-coslr-300e_in1k.py) | | 68.55 |
| DeepCluster | [deepcluster-sobel_resnet50_8xb64-steplr-200e_in1k.py](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/deepcluster/deepcluster-sobel_resnet50_8xb64-steplr-200e_in1k.py) | | 46.92 |
| DenseCL | [densecl_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/densecl/densecl_resnet50_8xb32-coslr-200e_in1k.py) | | 63.34 |
| MoCo v2 | [mocov2_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mocov2/mocov2_resnet50_8xb32-coslr-200e_in1k.py) | | 67.56 |
| NPID | [npid_resnet50_8xb32-steplr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/npid/npid_resnet50_8xb32-steplr-200e_in1k.py) | | 58.16 |
| ODC | [odc_resnet50_8xb64-steplr-440e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/odc/odc_resnet50_8xb64-steplr-440e_in1k.py) | | 53.42 |
| Relative Location | [relative-loc_resnet50_8xb64-steplr-70e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/relative_loc/relative-loc_resnet50_8xb64-steplr-70e_in1k.py) | | 39.65 |
| Rotation Prediction | [rotation-pred_resnet50_8xb16-steplr-70e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/rotation_pred/rotation-pred_resnet50_8xb16-steplr-70e_in1k.py) | | 44.35 |
| SimCLR | [simclr_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simclr/simclr_resnet50_8xb32-coslr-200e_in1k.py) | | 58.92 |
| SimSiam | [simsiam_resnet50_8xb32-coslr-100e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-100e_in1k.py) | SimSiam 论文设置 | 68.20 |
| | [simsiam_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-200e_in1k.py) | SimSiam 论文设置 | 69.80 |
| SwAV | [swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/swav/swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96.py) | SwAV 论文设置 | 70.55 |
| MoCo v3 | [mocov3_vit-small-p16_32xb128-fp16-coslr-300e_in1k-224](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mocov3/mocov3_vit-small-p16_32xb128-fp16-coslr-300e_in1k-224.py) | MoCo v3 论文设置 | 73.07 |
| DenseCL | [densecl_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/densecl/densecl_resnet50_8xb32-coslr-200e_in1k.py) | | 63.62 |
| MoCo v2 | [mocov2_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mocov2/mocov2_resnet50_8xb32-coslr-200e_in1k.py) | | 67.58 |
| NPID | [npid_resnet50_8xb32-steplr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/npid/npid_resnet50_8xb32-steplr-200e_in1k.py) | | 57.97 |
| ODC | [odc_resnet50_8xb64-steplr-440e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/odc/odc_resnet50_8xb64-steplr-440e_in1k.py) | | 53.43 |
| Relative Location | [relative-loc_resnet50_8xb64-steplr-70e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/relative_loc/relative-loc_resnet50_8xb64-steplr-70e_in1k.py) | | 38.78 |
| Rotation Prediction | [rotation-pred_resnet50_8xb16-steplr-70e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/rotation_pred/rotation-pred_resnet50_8xb16-steplr-70e_in1k.py) | | 48.12 |
| SimCLR | [simclr_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simclr/simclr_resnet50_8xb32-coslr-200e_in1k.py) | | 57.28 |
| SimSiam | [simsiam_resnet50_8xb32-coslr-100e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-100e_in1k.py) | SimSiam 论文设置 | 68.28 |
| | [simsiam_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-200e_in1k.py) | SimSiam 论文设置 | 69.84 |
| SwAV | [swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/swav/swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96.py) | SwAV 论文设置 | 70.47 |
| MoCo v3 | [mocov3_vit-small-p16_32xb128-fp16-coslr-300e_in1k-224](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mocov3/mocov3_vit-small-p16_32xb128-fp16-coslr-300e_in1k-224.py) | MoCo v3 论文设置 | 73.19 |
### ImageNet 微调
| 算法 | 配置文件 | 备注 | Top-1 (%) |
| ---- | ------------------------------------------------------------------------------------------------------------------------------------------------------------- | ---- | --------- |
| MAE | [mae_vit-base-p16_8xb512-coslr-400e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mae/mae_vit-base-p16_8xb512-coslr-400e_in1k.py) | | 83.1 |
### COCO17 目标检测
在 COCO17 数据集的目标检测任务中，我们选用 [MoCo](http://openaccess.thecvf.com/content_CVPR_2020/papers/He_Momentum_Contrast_for_Unsupervised_Visual_Representation_Learning_CVPR_2020_paper.pdf) 的评估设置，基于 Mask-RCNN 网络架构，下列结果通过同样的 [配置文件](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/mmdetection/coco/mask_rcnn_r50_fpn_mstrain_1x_coco.py) 训练得到。
@ -72,15 +79,15 @@
在 Pascal VOC12 Aug 分割任务中，我们选用 [MMSeg](https://github.com/open-mmlab/mmsegmentation) 的评估设置, 基于 FCN 网络架构, 下列结果通过同样的 [配置文件](https://github.com/open-mmlab/mmselfsup/blob/master/configs/benchmarks/mmsegmentation/voc12aug/fcn_r50-d8_512x512_20k_voc12aug.py) 训练得到。
| 算法 | 配置文件 | mIOU |
| ------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----- |
| BYOL | [byol_resnet50_8xb32-accum16-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/byol/byol_resnet50_8xb32-accum16-coslr-200e_in1k.py) | 67.16 |
| DenseCL | [densecl_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/densecl/densecl_resnet50_8xb32-coslr-200e_in1k.py) | 69.47 |
| MoCo v2 | [mocov2_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mocov2/mocov2_resnet50_8xb32-coslr-200e_in1k.py) | 67.55 |
| NPID | [npid_resnet50_8xb32-steplr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/npid/npid_resnet50_8xb32-steplr-200e_in1k.py) | 65.45 |
| Relative Location | [relative-loc_resnet50_8xb64-steplr-70e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/relative_loc/relative-loc_resnet50_8xb64-steplr-70e_in1k.py) | 63.49 |
| Rotation Prediction | [rotation-pred_resnet50_8xb16-steplr-70e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/rotation_pred/rotation-pred_resnet50_8xb16-steplr-70e_in1k.py) | 64.31 |
| SimCLR | [simclr_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simclr/simclr_resnet50_8xb32-coslr-200e_in1k.py) | 64.03 |
| SimSiam | [simsiam_resnet50_8xb32-coslr-100e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-100e_in1k.py) | 48.35 |
| | [simsiam_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-200e_in1k.py) | 46.27 |
| SwAV | [swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/swav/swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96.py) | 63.73 |
| 算法 | 配置文件 | mIOU |
| ------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----- |
| BYOL | [byol_resnet50_8xb32-accum16-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/byol/byol_resnet50_8xb32-accum16-coslr-200e_in1k.py) | 67.16 |
| DenseCL | [densecl_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/densecl/densecl_resnet50_8xb32-coslr-200e_in1k.py) | 69.47 |
| MoCo v2 | [mocov2_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/mocov2/mocov2_resnet50_8xb32-coslr-200e_in1k.py) | 67.55 |
| NPID | [npid_resnet50_8xb32-steplr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/npid/npid_resnet50_8xb32-steplr-200e_in1k.py) | 65.45 |
| Relative Location | [relative-loc_resnet50_8xb64-steplr-70e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/relative_loc/relative-loc_resnet50_8xb64-steplr-70e_in1k.py) | 63.49 |
| Rotation Prediction | [rotation-pred_resnet50_8xb16-steplr-70e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/rotation_pred/rotation-pred_resnet50_8xb16-steplr-70e_in1k.py) | 64.31 |
| SimCLR | [simclr_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simclr/simclr_resnet50_8xb32-coslr-200e_in1k.py) | 64.03 |
| SimSiam | [simsiam_resnet50_8xb32-coslr-100e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-100e_in1k.py) | 48.35 |
| | [simsiam_resnet50_8xb32-coslr-200e_in1k](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/simsiam/simsiam_resnet50_8xb32-coslr-200e_in1k.py) | 46.27 |
| SwAV | [swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96](https://github.com/open-mmlab/mmselfsup/blob/master/configs/selfsup/swav/swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96.py) | 63.73 |

347
docs/zh_cn/tutorials/0_config.md
View File

@ -1,189 +1,175 @@
# Tutorial 0: Learn about Configs
# 教程 0: 学习配置
MMSelfSup mainly uses python files as configs. The design of our configuration file system integrates modularity and inheritance, facilitating users to conduct various experiments. All configuration files are placed in the `configs` folder. If you wish to inspect the config file in summary, you may run `python tools/misc/print_config.py` to see the complete config.
MMSelfSup 主要使用python文件作为配置。我们设计的配置文件系统集成了模块化和继承性，方便用户实施各种实验。所有的配置文件都放在 `configs` 文件夹。如果你想概要地审视配置文件，你可以执行 `python tools/misc/print_config.py` 查看完整配置。
<!-- TOC -->
- [Tutorial 0: Learn about Configs](#tutorial-0-learn-about-configs)
- [Config File and Checkpoint Naming Convention](#config-file-and-checkpoint-naming-convention)
- [Algorithm information](#algorithm-information)
- [Module information](#module-information)
- [Training information](#training-information)
- [Data information](#data-information)
- [Config File Name Example](#config-file-name-example)
- [Checkpoint Naming Convention](#checkpoint-naming-convention)
- [Config File Structure](#config-file-structure)
- [Inherit and Modify Config File](#inherit-and-modify-config-file)
- [Use intermediate variables in configs](#use-intermediate-variables-in-configs)
- [Ignore some fields in the base configs](#ignore-some-fields-in-the-base-configs)
- [Use some fields in the base configs](#use-some-fields-in-the-base-configs)
- [Modify config through script arguments](#modify-config-through-script-arguments)
- [Import user-defined modules](#import-user-defined-modules)
- [教程 0: 学习配置](#教程-0-学习配置)
- [配置文件与检查点命名约定](#配置文件与检查点命名约定)
- [算法信息](#算法信息)
- [模块信息](#模块信息)
- [训练信息](#训练信息)
- [数据信息](#数据信息)
- [配置文件命名示例](#配置文件命名示例)
- [检查点命名约定](#检查点命名约定)
- [配置文件结构](#配置文件结构)
- [继承和修改配置文件](#继承和修改配置文件)
- [使用配置中的中间变量](#使用配置中的中间变量)
- [忽略基础配置中的字段](#忽略基础配置中的字段)
- [使用基础配置中的字段](#使用基础配置中的字段)
- [通过脚本参数修改配置](#通过脚本参数修改配置)
- [导入用户定义模块](#导入用户定义模块)
<!-- TOC -->
## Config File and Checkpoint Naming Convention
## 配置文件与检查点命名约定
We follow the below convention to name config files. Contributors are advised to follow the same style. The config file names are divided into four parts: algorithm info, module information, training information and data information. Logically, different parts are concatenated by underscores `'_'`, and words in the same part are concatenated by dashes `'-'`.
我们遵循下述约定来命名配置文件并建议贡献者也遵循该命名风格。配置文件名字被分成4部分：算法信息、模块信息、训练信息和数据信息。逻辑上，不同部分用下划线连接 `'_'`，同一部分中的单词使用破折线 `'-'` 连接。
```
{algorithm}_{module}_{training_info}_{data_info}.py
```
- `algorithm info`：Algorithm information includes algorithm name, such as simclr, mocov2, etc.;
- `module info`： Module information is used to represent some backbone, neck, head information;
- `training info`：Training information, some training schedule, including batch size, lr schedule, data augment and the like;
- `data info`：Data information, dataset name, input size and so on, such as imagenet, cifar, etc.;
- `algorithm info`：包含算法名字的算法信息，例如simclr，mocov2等；
- `module info`： 模块信息，用来表示一些 backbone，neck 和 head 信息；
- `training info`：训练信息，即一些训练调度，包括批大小，学习率调度，数据增强等；
- `data info`：数据信息：数据集名字，输入大小等，例如 imagenet，cifar 等。
### Algorithm information
### 算法信息
```
{algorithm}-{misc}
```
`Algorithm` means the abbreviation from the paper and its version. E.g:
- `relative-loc` : The different word is concatenated by dashes `'-'`
`Algorithm` 表示论文中的算法缩写和版本。例如：
- `relative-loc`：不同单词之间使用破折线连接 `'-'`
- `simclr`
- `mocov2`
`misc` offers some other algorithm related information. E.g.
`misc` 提供一些其他算法相关信息。例如：
- `npid-ensure-neg`
- `deepcluster-sobel`
### Module information
### 模块信息
```
{backbone setting}_{neck setting}_{head_setting}
{backbone setting}-{neck setting}-{head_setting}
```
Here we use `'_'` to concatenate to make the name more readable.
The module information mainly includes the backbone information. E.g:
模块信息主要包含 backboe 信息。例如：
- `resnet50`
- `vit`（will be used in mocov3）
- `vit`（将会用在mocov3中）
Or there are some special settings which is needed to be mentioned in the config name. E.g:
- `resnet50-nofrz`: In some downstream tasks，the backbone will not froze stages while training
- `simclr_resnet50_mocov2-neck`: In simclr, using mocov2-neck to train, not simclr its own neck
或者其他一些需要在配置名字中强调的特殊的设置。例如：
- `resnet50-nofrz`：在一些下游任务的训练中，该 backbone 不会冻结 stages
### Training information
### 训练信息
Training related settings，including batch size, lr schedule, data augment, etc.
- Batch size, the format is `{gpu x batch_per_gpu}`，like `8xb32`;
- Training recipe，the methods will be arranged in the order `{pipeline aug}-{train aug}-{loss trick}-{scheduler}-{epochs}`.
训练相关的配置，包括 batch size, lr schedule, data augment 等。
- Batch size，格式是 `{gpu x batch_per_gpu}` ，例如 `8xb32`；
- Training recipe，该方法以如下顺序组织：`{pipeline aug}-{train aug}-{loss trick}-{scheduler}-{epochs}`
E.g:
- `8xb32-mcrop-2-6-coslr-200e` : `mcrop` is proposed in SwAV named multi-crop，part of pipeline. 2 and 6 means that 2 pipelines will output 2 and 6 crops correspondingly，the crop size is recorded in data information;
- `8xb32-accum16-coslr-200e` : `accum16` means the gradient will accumulate for 16 iterations，then the weights will be updated.
例如：
- `8xb32-mcrop-2-6-coslr-200e`：`mcrop` 是 SwAV 提出的 pipeline 中的名为 multi-crop 的一部分。2 和 6 表示 2 个 pipeline 分别输出 2 个和 6 个裁剪图，而且裁剪信息记录在数据信息中；
- `8xb32-accum16-coslr-200e`：`accum16` 表示权重会在梯度累积16个迭代之后更新。
### Data information
Data information contains the dataset, input size, etc. E.g:
- `in1k` : `ImageNet1k` dataset, default to use the input image size of 224x224
- `in1k-384px` : Indicates that the input image size is 384x384
### 数据信息
数据信息包含数据集，输入大小等。例如：
- `in1k`：`ImageNet1k` 数据集，默认使用的输入图像大小是 224x224
- `in1k-384px`：表示输入图像大小是384x384
- `cifar10`
- `inat18` : `iNaturalist2018` dataset，it has 8142 classes
- `inat18`：`iNaturalist2018` 数据集，包含 8142 类
- `places205`
### Config File Name Example
```
mocov2_resnet50_simclr-neck_8xb32-coslr-200e_in1k.py
```
- `mocov2`: Algorithm information
- `resnet50_simclr-neck`: Module information
- `resnet50`: Backbone
- `simclr-neck`: Using special neck，not the default mocov2-neck
- `8xb32-coslr-200e`: Training information
- `8xb32`: Use 8 GPUs in total，and the batch size is 32 per GPU
- `coslr`: Use cosine learning rate scheduler
- `200e`: Train the model for 200 epoch
- `in1k`: Data information, train on ImageNet1k dataset
### 配置文件命名示例
```
swav_resnet50_8xb32-mcrop-2-6-coslr-200e_in1k-224-96.py
```
- `swav`: Algorithm information
- `resnet50`: Module information
- `8xb32-mcrop-2-6-coslr-200e`: Training information
- `8xb32`: Use 8 GPUs in total，and the batch size is 32 per GPU
- `mcrop-2-6`:Use multi-crop data augment method
- `coslr`: Use cosine learning rate scheduler
- `200e`: Train the model for 200 epoch
- `in1k-224-96`: Data information，train on ImageNet1k dataset，the input sizes are 224x224 and 96x96
- `swav`：算法信息
- `resnet50`：模块信息
- `8xb32-mcrop-2-6-coslr-200e`：训练信息
- `8xb32`：共使用 8 张 GPU，每张 GPU 上的 batch size 是 32
- `mcrop-2-6`：使用 multi-crop 数据增强方法
- `coslr`：使用余弦学习率调度器
- `200e`：训练模型200个周期
- `in1k-224-96`：数据信息，在 ImageNet1k 数据集上训练，输入大小是 224x224 和 96x96
### Checkpoint Naming Convention
### 检查点命名约定
The naming of the weight mainly includes the configuration file name, date and hash value.
权重的命名主要包括配置文件名字，日期和哈希值。
```
{config_name}_{date}-{hash}.pth
```
## Config File Structure
## 配置文件结构
There are four kinds of basic component file in the `configs/_base_` folders, namely：
在 `configs/_base_` 文件中，有 4 种类型的基础组件文件，即
- models
- datasets
- schedules
- runtime
You can easily build your own training config file by inherit some base config files. And the configs that are composed by components from `_base_` are called _primitive_.
你可以通过继承一些基础配置文件快捷地构建你自己的配置。由 `_base_` 下的组件组成的配置被称为 _原始配置（primitive）_。
For easy understanding, we use MoCo v2 as a example and comment the meaning of each line. For more detaile, please refer to the API documentation.
为了易于理解，我们使用 MoCo v2 作为一个例子，并对它的每一行做出注释。若想了解更多细节，请参考 API 文档。
The config file `configs/selfsup/mocov2/mocov2_resnet50_8xb32-coslr-200e_in1k.py` is displayed below.
配置文件 `configs/selfsup/mocov2/mocov2_resnet50_8xb32-coslr-200e_in1k.py` 如下所述。
```python
_base_ = [
'../_base_/models/mocov2.py', # model
'../_base_/datasets/imagenet_mocov2.py', # data
'../_base_/schedules/sgd_coslr-200e_in1k.py', # training schedule
'../_base_/default_runtime.py', # runtime setting
'../_base_/models/mocov2.py', # 模型
'../_base_/datasets/imagenet_mocov2.py', # 数据
'../_base_/schedules/sgd_coslr-200e_in1k.py', # 训练调度
'../_base_/default_runtime.py', # 运行时设置
]
# Here we inherit runtime settings and modify the max_keep_ckpts.
# the max_keep_ckpts controls the max number of ckpt file in your work_dirs
# if it is 3, when CheckpointHook (in mmcv) saves the 4th ckpt
# it will remove the oldest one to keep the number of total ckpts as 3
# 在这里，我们继承运行时设置并修改 max_keep_ckpts。
# max_keep_ckpts 控制在你的 work_dirs 中最大的ckpt文件的数量
# 如果它是3，当 CheckpointHook (在mmcv中) 保存第 4 个 ckpt 时，
# 它会移除最早的那个，使总的 ckpt 文件个数保持为 3
checkpoint_config = dict(interval=10, max_keep_ckpts=3)
```
```{note}
The 'type' in the configuration file is not a constructed parameter, but a class name.
配置文件中的 'type' 是一个类名，而不是参数的一部分。
```
`../_base_/models/mocov2.py` is the base model config for MoCo v2.
`../_base_/models/mocov2.py` 是 MoCo v2 的基础模型配置。
```python
model = dict(
type='MoCo', # Algorithm name
queue_len=65536, # Number of negative keys maintained in the queue
feat_dim=128, # Dimension of compact feature vectors, equal to the out_channels of the neck
momentum=0.999, # Momentum coefficient for the momentum-updated encoder
type='MoCo', # 算法名字
queue_len=65536, # 队列中维护的负样本数量
feat_dim=128, # 紧凑特征向量的维度，等于 neck 的 out_channels
momentum=0.999, # 动量更新编码器的动量系数
backbone=dict(
type='ResNet', # Backbone name
depth=50, # Depth of backbone, ResNet has options of 18, 34, 50, 101, 152
in_channels=3, # The channel number of the input images
out_indices=[4], # The output index of the output feature maps, 0 for conv-1, x for stage-x
norm_cfg=dict(type='BN')), # Dictionary to construct and config norm layer
depth=50, # backbone 深度，ResNet 可以选择 18、34、50、101、 152
in_channels=3, # 输入图像的通道数
out_indices=[4], # 输出特征图的输出索引，0 表示 conv-1，x 表示 stage-x
norm_cfg=dict(type='BN')), # 构建一个字典并配置 norm 层
neck=dict(
type='MoCoV2Neck', # Neck name
in_channels=2048, # Number of input channels
hid_channels=2048, # Number of hidden channels
out_channels=128, # Number of output channels
with_avg_pool=True), # Whether to apply the global average pooling after backbone
in_channels=2048, # 输入通道数
hid_channels=2048, # 隐层通道数
out_channels=128, # 输出通道数
with_avg_pool=True), # 是否在 backbone 之后使用全局平均池化
head=dict(
type='ContrastiveHead', # Head name, indicates that the MoCo v2 use contrastive loss
temperature=0.2)) # The temperature hyper-parameter that controls the concentration level of the distribution.
type='ContrastiveHead', # Head name, 表示 MoCo v2 使用 contrastive loss
temperature=0.2)) # 控制分布聚集程度的温度超参数
```
`../_base_/datasets/imagenet_mocov2.py` is the base dataset config for MoCo v2.
`../_base_/datasets/imagenet_mocov2.py` 是 MoCo v2 的基础数据集配置。
```python
# dataset settings
data_source = 'ImageNet' # data source name
dataset_type = 'MultiViewDataset' # dataset type is related to the pipeline composing
# 数据集配置
data_source = 'ImageNet' # 数据源名字
dataset_type = 'MultiViewDataset' # 组成 pipeline 的数据集类型
img_norm_cfg = dict(
mean=[0.485, 0.456, 0.406], # Mean values used to pre-training the pre-trained backbone models
std=[0.229, 0.224, 0.225]) # Standard variance used to pre-training the pre-trained backbone models
# The difference between mocov2 and mocov1 is the transforms in the pipeline
mean=[0.485, 0.456, 0.406], # 用来预训练预训练 backboe 模型的均值
std=[0.229, 0.224, 0.225]) # 用来预训练预训练 backbone 模型的标准差
# mocov2 和 mocov1 之间的差异在于 pipeline 中的 transforms
train_pipeline = [
dict(type='RandomResizedCrop', size=224, scale=(0.2, 1.)), # RandomResizedCrop
dict(
type='RandomAppliedTrans', # Random apply ColorJitter augment method with probability 0.8
type='RandomAppliedTrans', # 以0.8的概率随机使用 ColorJitter 增强方法
transforms=[
dict(
type='ColorJitter',
@ -193,89 +179,89 @@ train_pipeline = [
hue=0.1)
],
p=0.8),
dict(type='RandomGrayscale', p=0.2), # RandomGrayscale with probability 0.2
dict(type='GaussianBlur', sigma_min=0.1, sigma_max=2.0, p=0.5), # Random GaussianBlur with probability 0.5
dict(type='RandomHorizontalFlip'), # Randomly flip the picture horizontally
dict(type='RandomGrayscale', p=0.2), # 0.2概率的 RandomGrayscale
dict(type='GaussianBlur', sigma_min=0.1, sigma_max=2.0, p=0.5), # 0.5概率的随机 GaussianBlur
dict(type='RandomHorizontalFlip'), # 随机水平翻转图像
]
# prefetch
prefetch = False # Whether to using prefetch to speed up the pipeline
prefetch = False # 是否使用 prefetch 加速 pipeline
if not prefetch:
train_pipeline.extend(
[dict(type='ToTensor'),
dict(type='Normalize', **img_norm_cfg)])
# dataset summary
# 数据集汇总
data = dict(
samples_per_gpu=32, # Batch size of a single GPU, total 32*8=256
workers_per_gpu=4, # Worker to pre-fetch data for each single GPU
drop_last=True, # Whether to drop the last batch of data
samples_per_gpu=32, # 单张 GPU 的批大小, 共 32*8=256
workers_per_gpu=4, # 每张 GPU 用来 pre-fetch 数据的 worker 个数
drop_last=True, # 是否丢弃最后一个 batch 的数据
train=dict(
type=dataset_type, # dataset name
type=dataset_type, # 数据集名字
data_source=dict(
type=data_source, # data source name
data_prefix='data/imagenet/train', # Dataset root, when ann_file does not exist, the category information is automatically obtained from the root folder
ann_file='data/imagenet/meta/train.txt', # ann_file existes, the category information is obtained from file
type=data_source, # 数据源名字
data_prefix='data/imagenet/train', # 数据集根目录, 当 ann_file 不存在时，类别信息自动从该根目录自动获取
ann_file='data/imagenet/meta/train.txt', # 若 ann_file 存在，类别信息从该文件获取
),
num_views=[2], # The number of different views from pipeline
pipelines=[train_pipeline], # The train pipeline
prefetch=prefetch, # The boolean value
num_views=[2], # pipeline 中不同的视图个数
pipelines=[train_pipeline], # 训练 pipeline
prefetch=prefetch, # 布尔值
))
```
`../_base_/schedules/sgd_coslr-200e_in1k.py` is the base schedule config for MoCo v2.
`../_base_/schedules/sgd_coslr-200e_in1k.py` 是 MoCo v2 的基础调度配置。
```python
# optimizer
# 优化器
optimizer = dict(
type='SGD', # Optimizer type
lr=0.03, # Learning rate of optimizers, see detail usages of the parameters in the documentation of PyTorch
weight_decay=1e-4, # Momentum parameter
momentum=0.9) # Weight decay of SGD
# Config used to build the optimizer hook, refer to https://github.com/open-mmlab/mmcv/blob/master/mmcv/runner/hooks/optimizer.py#L8 for implementation details.
optimizer_config = dict() # this config can set grad_clip, coalesce, bucket_size_mb, etc.
type='SGD', # 优化器类型
lr=0.03, # 优化器的学习率, 参数的详细使用请参阅 PyTorch 文档
weight_decay=1e-4, # 动量参数
momentum=0.9) # SGD 的权重衰减
# 用来构建优化器钩子的配置，请参考 https://github.com/open-mmlab/mmcv/blob/master/mmcv/runner/hooks/optimizer.py#L8 中的实现细节。
optimizer_config = dict() # 这个配置可以设置 grad_clip，coalesce，bucket_size_mb 等。
# learning policy
# Learning rate scheduler config used to register LrUpdater hook
# 学习策略
# 用来注册 LrUpdater 钩子的学习率调度配置
lr_config = dict(
policy='CosineAnnealing', # The policy of scheduler, also support Step, Cyclic, etc. Refer to details of supported LrUpdater from https://github.com/open-mmlab/mmcv/blob/master/mmcv/runner/hooks/lr_updater.py#L9.
min_lr=0.) # The minimum lr setting in CosineAnnealing
policy='CosineAnnealing', # 调度器策略，也支持 Step，Cyclic 等。 LrUpdater 支持的细节请参考 https://github.com/open-mmlab/mmcv/blob/master/mmcv/runner/hooks/lr_updater.py#L9。
min_lr=0.) # CosineAnnealing 中的最小学习率设置
# runtime settings
# 运行时设置
runner = dict(
type='EpochBasedRunner', # Type of runner to use (i.e. IterBasedRunner or EpochBasedRunner)
max_epochs=200) # Runner that runs the workflow in total max_epochs. For IterBasedRunner use `max_iters`
type='EpochBasedRunner', # 使用的 runner 的类型 (例如 IterBasedRunner 或 EpochBasedRunner)
max_epochs=200) # 运行工作流周期总数的 Runner 的 max_epochs，对于IterBasedRunner 使用 `max_iters`
```
`../_base_/default_runtime.py` is the default runtime settings.
`../_base_/default_runtime.py` 是运行时的默认配置。
```python
# checkpoint saving
checkpoint_config = dict(interval=10) # The save interval is 1
# 保存检查点
checkpoint_config = dict(interval=10) # 保存间隔是10
# yapf:disable
log_config = dict(
interval=50, # Interval to print the log
interval=50, # 打印日志的间隔
hooks=[
dict(type='TextLoggerHook'), # The Tensorboard logger is also supported
dict(type='TextLoggerHook'), # 也支持 Tensorboard logger
# dict(type='TensorboardLoggerHook'),
])
# yapf:enable
# runtime settings
dist_params = dict(backend='nccl') # Parameters to setup distributed training, the port can also be set.
log_level = 'INFO' # The output level of the log.
load_from = None # Runner to load ckpt
resume_from = None # Resume checkpoints from a given path, the training will be resumed from the epoch when the checkpoint's is saved.
workflow = [('train', 1)] # Workflow for runner. [('train', 1)] means there is only one workflow and the workflow named 'train' is executed once.
persistent_workers = True # The boolean type to set persistent_workers in Dataloader. see detail in the documentation of PyTorch
# 运行时设置
dist_params = dict(backend='nccl') # 设置分布式训练的参数，端口也支持设置。
log_level = 'INFO' # 日志的输出 level。
load_from = None # 加载 ckpt
resume_from = None # 从给定的路径恢复检查点，将会从检查点保存时的周期恢复训练。
workflow = [('train', 1)] # Workflow for runner. [('train', 1)] 表示有一个 workflow，该 workflow 名字是 'train' 且执行一次。
persistent_workers = True # Dataloader 中设置 persistent_workers 的布尔值，详细信息请参考 PyTorch 文档
```
## Inherit and Modify Config File
## 继承和修改配置文件
For easy understanding, we recommend contributors to inherit from existing methods.
为了易于理解，我们推荐贡献者从现有方法继承。
For all configs under the same folder, it is recommended to have only **one** _primitive_ config. All other configs should inherit from the _primitive_ config. In this way, the maximum of inheritance level is 3.
对于同一个文件夹下的所有配置，我们推荐只使用**一个** _原始（primitive）_ 配置。其他所有配置应当从 _原始（primitive）_ 配置继承，这样最大的继承层次为 3。
For example, if your config file is based on MoCo v2 with some other modification, you can first inherit the basic MoCo v2 structure, dataset and other training setting by specifying `_base_ ='./mocov2_resnet50_8xb32-coslr-200e_in1k.py.py'` (The path relative to your config file), and then modify the necessary parameters in the config file. A more specific example, now we want to use almost all configs in `configs/selfsup/mocov2/mocov2_resnet50_8xb32-coslr-200e_in1k.py.py`, but change the number of training epochs from 200 to 800, modify when to decay the learning rate, and modify the dataset path, you can create a new config file `configs/selfsup/mocov2/mocov2_resnet50_8xb32-coslr-800e_in1k.py.py` with content as below:
例如，如果你的配置文件是基于 MoCo v2 做一些修改，首先你可以通过指定 `_base_ ='./mocov2_resnet50_8xb32-coslr-200e_in1k.py.py'` （相对于你的配置文件的路径）继承基本的 MoCo v2 结构，数据集和其他训练设置，接着在配置文件中修改一些必要的参数。现在，我们举一个更具体的例子，我们想使用 `configs/selfsup/mocov2/mocov2_resnet50_8xb32-coslr-200e_in1k.py.py` 中几乎所有的配置，但是将训练周期数从 200 修改为 800，修改学习率衰减的时机和数据集路径，你可以创建一个名为 `configs/selfsup/mocov2/mocov2_resnet50_8xb32-coslr-800e_in1k.py.py` 的新配置文件，内容如下：
```python
_base_ = './mocov2_resnet50_8xb32-coslr-200e_in1k.py'
@ -283,11 +269,11 @@ _base_ = './mocov2_resnet50_8xb32-coslr-200e_in1k.py'
runner = dict(max_epochs=800)
```
### Use intermediate variables in configs
### 使用配置中的中间变量
Some intermediate variables are used in the configuration file. The intermediate variables make the configuration file clearer and easier to modify.
在配置文件中使用一些中间变量会使配置文件更加清晰和易于修改。
For example, `data_source`, `dataset_type`, `train_pipeline`, `prefetch` are the intermediate variables of the data. We first need to define them and then pass them to `data`.
例如：数据中的中间变量有 `data_source`, `dataset_type`, `train_pipeline`, `prefetch`. 我们先定义它们再将它们传进 `data`。
```python
data_source = 'ImageNet'
@ -296,7 +282,7 @@ img_norm_cfg = dict(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
train_pipeline = [...]
# prefetch
prefetch = False # Whether to using prefetch to speed up the pipeline
prefetch = False # 是否使用 prefetch 加速 pipeline
if not prefetch:
train_pipeline.extend(
[dict(type='ToTensor'),
@ -315,11 +301,11 @@ data = dict(
```
### Ignore some fields in the base configs
### 忽略基础配置中的字段
Sometimes, you need to set `_delete_=True` to ignore some domain content in the basic configuration file. You can refer to [mmcv](https://mmcv.readthedocs.io/en/latest/understand_mmcv/config.html#inherit-from-base-config-with-ignored-fields) for more instructions.
有时候，你需要设置 `_delete_=True` 来忽略基础配置文件中一些域的内容。 你可以参考 [mmcv](https://mmcv.readthedocs.io/zh_CN/latest/understand_mmcv/config.html#inherit-from-base-config-with-ignored-fields) 获得更多说明。
The following is an example. If you want to use `MoCoV2Neck` in simclr setting, just using inheritance and directly modify it will report `get unexcepected keyword 'num_layers'` error, because the `'num_layers'` field of the basic config in `model.neck` domain information is reserved, and you need to add `_delete_=True` to ignore the content of `model.neck` related fields in the basic configuration file:
接下来是一个例子。如果你希望在 simclr 的设置中使用 `MoCoV2Neck`，仅仅继承并直接修改将会报 `get unexcepected keyword 'num_layers'` 错误，因为在 `model.neck` 域信息中，基础配置 `'num_layers'` 字段被保存下来了， 你需要添加 `_delete_=True` 来忽略 `model.neck` 在基础配置文件中的有关字段的内容。
```python
_base_ = 'simclr_resnet50_8xb32-coslr-200e_in1k.py'
@ -333,11 +319,11 @@ model = dict(
out_channels=128,
with_avg_pool=True))
```
### Use some fields in the base configs
### 使用基础配置中的字段
Sometimes, you may refer to some fields in the `_base_` config, so as to avoid duplication of definitions. You can refer to [mmcv](https://mmcv.readthedocs.io/en/latest/understand_mmcv/config.html#reference-variables-from-base) for some more instructions.
有时候，你可能引用 `_base_` 配置中一些字段，以避免重复定义。你可以参考[mmcv](https://mmcv.readthedocs.io/zh_CN/latest/understand_mmcv/config.html#reference-variables-from-base) 获取更多的说明。
The following is an example of using auto augment in the training data preprocessing pipeline， refer to `configs/selfsup/odc/odc_resnet50_8xb64-steplr-440e_in1k.py`. When defining `num_classes`, just add the definition file name of auto augment to `_base_`, and then use `{{_base_.num_classes}}` to reference the variables:
下面是在训练数据预处理 pipeline 中使用 auto augment 的一个例子，请参考 `configs/selfsup/odc/odc_resnet50_8xb64-steplr-440e_in1k.py`。当定义 `num_classes` 时，只需要将 auto augment 的定义文件名添入到 `_base_`，并使用 `{{_base_.num_classes}}` 来引用这些变量：
```python
_base_ = [
@ -366,43 +352,38 @@ lr_config = dict(policy='step', step=[400], gamma=0.4)
# runtime settings
runner = dict(type='EpochBasedRunner', max_epochs=440)
# the max_keep_ckpts controls the max number of ckpt file in your work_dirs
# if it is 3, when CheckpointHook (in mmcv) saves the 4th ckpt
# it will remove the oldest one to keep the number of total ckpts as 3
# max_keep_ckpts 控制在你的 work_dirs 中保存的 ckpt 的最大数目
# 如果它等于3，CheckpointHook（在mmcv中）在保存第 4 个 ckpt 时，
# 它会移除最早的那个，使总的 ckpt 文件个数保持为 3
checkpoint_config = dict(interval=10, max_keep_ckpts=3)
```
## Modify config through script arguments
## 通过脚本参数修改配置
When users use the script "tools/train.py" or "tools/test.py" to submit tasks or use some other tools, they can directly modify the content of the configuration file used by specifying the `--cfg-options` parameter.
当用户使用脚本 "tools/train.py" 或 "tools/test.py" 提交任务，或者其他工具时，可以通过指定 `--cfg-options` 参数来直接修改配置文件中内容。
- Update config keys of dict chains.
- 更新字典链中的配置的键
The config options can be specified following the order of the dict keys in the original config.
For example, `--cfg-options model.backbone.norm_eval=False` changes the all BN modules in model backbones to `train` mode.
配置项可以通过遵循原始配置中键的层次顺序指定。例如，`--cfg-options model.backbone.norm_eval=False` 改变模型 backbones 中的所有 BN 模块为 `train` 模式。
- Update keys inside a list of configs.
- 更新列表中配置的键
Some config dicts are composed as a list in your config. For example, the training pipeline `data.train.pipeline` is normally a list
e.g. `[dict(type='LoadImageFromFile'), dict(type='TopDownRandomFlip', flip_prob=0.5), ...]`. If you want to change `'flip_prob=0.5'` to `'flip_prob=0.0'` in the pipeline,
you may specify `--cfg-options data.train.pipeline.1.flip_prob=0.0`.
你的配置中的一些配置字典是由列表组成。例如，训练 pipeline `data.train.pipeline` 通常是一个列表。例如 `[dict(type='LoadImageFromFile'), dict(type='TopDownRandomFlip', flip_prob=0.5), ...]`。如果你想要在 pipeline 中将 `'flip_prob=0.5'` 修改为 `'flip_prob=0.0'`，你可以指定 `--cfg-options data.train.pipeline.1.flip_prob=0.0`
- Update values of list/tuples.
- 更新 list/tuples 中的值
If the value to be updated is a list or a tuple. For example, the config file normally sets `workflow=[('train', 1)]`. If you want to
change this key, you may specify `--cfg-options workflow="[(train,1),(val,1)]"`. Note that the quotation mark \" is necessary to
support list/tuple data types, and that **NO** white space is allowed inside the quotation marks in the specified value.
如果想要更新的值是一个列表或者元组，例如：配置文件通常设置 `workflow=[('train', 1)]`。如果你想要改变这个键，你可以指定 `--cfg-options workflow="[(train,1),(val,1)]"`。注意：对于 list/tuple 数据类型，引号\" 是必须的，并且在指定值的时候，在引号中 **NO** 空白字符。
## Import user-defined modules
## 导入用户定义模块
```{note}
This part may only be used when using other MM-codebase, like mmcls as a third party library to build your own project, and beginners can skip it.
这部分内容初学者可以跳过，只在使用其他 MM-codebase 时会用到，例如使用 mmcls 作为第三方库来构建你的工程。
```
You may use other MM-codebase to complete your project and create new classes of datasets, models, data enhancements, etc. in the project. In order to streamline the code, you can use MM-codebase as a third-party library, you just need to keep your own extra code and import your own custom module in the configuration files. For examples, you may refer to [OpenMMLab Algorithm Competition Project](https://github.com/zhangrui-wolf/openmmlab-competition-2021) .
你可能使用其他的 MM-codebase 来完成你的工程，并在工程中创建新的数据集类，模型类，数据增强类等。为了简化代码，你可以使用 MM-codebase 作为第三方库，只需要保存你自己额外的代码，并在配置文件中导入自定义模块。你可以参考 [OpenMMLab Algorithm Competition Project](https://github.com/zhangrui-wolf/openmmlab-competition-2021) 中的例子。
Add the following code to your own configuration files:
在你自己的配置文件中添加如下所述的代码：
```python
custom_imports = dict(

3
mmselfsup/core/hooks/__init__.py
View File

@ -1,4 +1,5 @@
# Copyright (c) OpenMMLab. All rights reserved.
from .cosineAnnealing_hook import StepFixCosineAnnealingLrUpdaterHook
from .deepcluster_hook import DeepClusterHook
from .densecl_hook import DenseCLHook
from .momentum_update_hook import MomentumUpdateHook
@ -10,5 +11,5 @@ from .swav_hook import SwAVHook
__all__ = [
'MomentumUpdateHook', 'DeepClusterHook', 'DenseCLHook', 'ODCHook',
'DistOptimizerHook', 'GradAccumFp16OptimizerHook', 'SimSiamHook',
'SwAVHook'
'SwAVHook', 'StepFixCosineAnnealingLrUpdaterHook'
]

35
mmselfsup/core/hooks/cosineAnnealing_hook.py 100644
View File

@ -0,0 +1,35 @@
# Copyright (c) OpenMMLab. All rights reserved.
from mmcv.runner import HOOKS
from mmcv.runner.hooks.lr_updater import (CosineAnnealingLrUpdaterHook,
annealing_cos)
@HOOKS.register_module()
class StepFixCosineAnnealingLrUpdaterHook(CosineAnnealingLrUpdaterHook):
def get_lr(self, runner, base_lr):
if self.by_epoch:
progress = runner.epoch
max_progress = runner.max_epochs
# Delete warmup epochs
if self.warmup is not None:
progress = progress - self.warmup_iters // len(
runner.data_loader)
max_progress = max_progress - self.warmup_iters // len(
runner.data_loader)
else:
progress = runner.iter
max_progress = runner.max_iters
# Delete warmup iters
if self.warmup is not None:
progress = progress - self.warmup_iters
max_progress = max_progress - self.warmup_iters
if self.min_lr_ratio is not None:
target_lr = base_lr * self.min_lr_ratio
else:
target_lr = self.min_lr
return annealing_cos(base_lr, target_lr, progress / max_progress)

6
mmselfsup/core/optimizer/__init__.py
View File

@ -1,6 +1,10 @@
# Copyright (c) OpenMMLab. All rights reserved.
from .builder import build_optimizer
from .constructor import DefaultOptimizerConstructor
from .mae_finetune_constructor import MAEFtOptimizerConstructor
from .optimizers import LARS
__all__ = ['LARS', 'build_optimizer', 'DefaultOptimizerConstructor']
__all__ = [
'LARS', 'build_optimizer', 'DefaultOptimizerConstructor',
'MAEFtOptimizerConstructor'
]

3
mmselfsup/core/optimizer/constructor.py
View File

@ -22,7 +22,7 @@ class DefaultOptimizerConstructor:
- any arguments of the corresponding optimizer type, e.g.,
lr, weight_decay, momentum, etc.
paramwise_cfg (dict, optional): Parameter-wise options.
Defaults to None
Defaults to None.
Example 1:
>>> model = torch.nn.modules.Conv1d(1, 1, 1)
@ -47,6 +47,7 @@ class DefaultOptimizerConstructor:
model = model.module
optimizer_cfg = self.optimizer_cfg.copy()
paramwise_options = self.paramwise_cfg
# if no paramwise option is specified, just use the global setting
if paramwise_options is None:
optimizer_cfg['params'] = model.parameters()

118
mmselfsup/core/optimizer/mae_finetune_constructor.py 100644
View File

@ -0,0 +1,118 @@
# Copyright (c) OpenMMLab. All rights reserved.
import re
import torch.distributed as dist
from mmcv.runner.optimizer.builder import OPTIMIZER_BUILDERS, OPTIMIZERS
from mmcv.utils import build_from_cfg, print_log
@OPTIMIZER_BUILDERS.register_module()
class MAEFtOptimizerConstructor:
"""Rewrote default constructor for optimizers. By default each parameter
share the same optimizer settings, and we provide an argument
``paramwise_cfg`` to specify parameter-wise settings and set layer-wise
learning rate. It is a dict and may contain the following fields:
Args:
model (:obj:`nn.Module`): The model with parameters to be optimized.
optimizer_cfg (dict): The config dict of the optimizer.
Positional fields are
- `type`: class name of the optimizer.
Optional fields are
- any arguments of the corresponding optimizer type, e.g.,
lr, weight_decay, momentum, etc.
paramwise_cfg (dict, optional): Parameter-wise options.
Defaults to None
layer_decay (float): base value for layer wise learning rate decay.
Defaults to 0.0
Example 1:
>>> model = torch.nn.modules.Conv1d(1, 1, 1)
>>> optimizer_cfg = dict(type='SGD', lr=0.01, momentum=0.9,
>>> weight_decay=0.0001)
>>> paramwise_cfg = dict('bias': dict(weight_decay=0., \
lars_exclude=True))
>>> optim_builder = DefaultOptimizerConstructor(
>>> optimizer_cfg, paramwise_cfg)
>>> optimizer = optim_builder(model)
"""
def __init__(self, optimizer_cfg, paramwise_cfg=None):
if not isinstance(optimizer_cfg, dict):
raise TypeError('optimizer_cfg should be a dict',
f'but got {type(optimizer_cfg)}')
self.optimizer_cfg = optimizer_cfg
self.paramwise_cfg = {} if paramwise_cfg is None else paramwise_cfg
self.layer_decay = self.optimizer_cfg.pop('layer_decay', 0.0)
def __call__(self, model):
if hasattr(model, 'module'):
model = model.module
optimizer_cfg = self.optimizer_cfg.copy()
paramwise_options = self.paramwise_cfg
# generate layer-wise lr decay
if self.layer_decay > 0:
self._generate_layer_wise_lr_decay(model, paramwise_options)
# if no paramwise option is specified, just use the global setting
if paramwise_options is None:
optimizer_cfg['params'] = model.parameters()
return build_from_cfg(optimizer_cfg, OPTIMIZERS)
else:
assert isinstance(paramwise_options, dict)
params = []
for name, param in model.named_parameters():
param_group = {'params': [param]}
if not param.requires_grad:
params.append(param_group)
continue
for regexp, options in paramwise_options.items():
if re.search(regexp, name):
for key, value in options.items():
if key.endswith('_mult'): # is a multiplier
key = key[:-5]
assert key in optimizer_cfg, \
f'{key} not in optimizer_cfg'
value = optimizer_cfg[key] * value
param_group[key] = value
if not dist.is_initialized() or \
dist.get_rank() == 0:
print_log(f'paramwise_options -- \
{name}: {key}={value}')
# otherwise use the global settings
params.append(param_group)
optimizer_cfg['params'] = params
return build_from_cfg(optimizer_cfg, OPTIMIZERS)
def _generate_layer_wise_lr_decay(self, model, paramwise_options):
"""Currently, we follow the same layer-wise lr decay schedule as
MAE."""
num_layers = len(model.backbone.layers) + 1
layer_scales = list(self.layer_decay**(num_layers - i)
for i in range(num_layers + 1))
if 'pos_embed' in paramwise_options:
paramwise_options['pos_embed'].update(
dict(lr_mult=layer_scales[0]))
else:
paramwise_options['pos_embed'] = dict(lr_mult=layer_scales[0])
if 'cls_token' in paramwise_options:
paramwise_options['cls_token'].update(
dict(lr_mult=layer_scales[0]))
else:
paramwise_options['cls_token'] = dict(lr_mult=layer_scales[0])
if 'patch_embed' in paramwise_options:
paramwise_options['patch_embed'].update(
dict(lr_mult=layer_scales[0]))
else:
paramwise_options['patch_embed'] = dict(lr_mult=layer_scales[0])
for i in range(num_layers - 1):
paramwise_options[f'backbone\\.layers\\.{i}\\.'] = dict(
lr_mult=layer_scales[i + 1])

2
mmselfsup/datasets/builder.py
View File

@ -131,7 +131,6 @@ def build_dataloader(dataset,
img_norm_cfg = kwargs.pop('img_norm_cfg')
else:
prefetch = False
data_loader = DataLoader(
dataset,
batch_size=batch_size,
@ -166,3 +165,4 @@ def worker_init_fn(worker_id, num_workers, rank, seed):
worker_seed = num_workers * rank + worker_id + seed
np.random.seed(worker_seed)
random.seed(worker_seed)
torch.manual_seed(worker_seed)

12
mmselfsup/datasets/data_sources/base.py
View File

@ -19,6 +19,8 @@ class BaseDataSource(object, metaclass=ABCMeta):
test_mode (bool): in train mode or test mode. Defaults to False.
color_type (str): The flag argument for :func:`mmcv.imfrombytes()`.
Defaults to color.
channel_order (str): The channel order of images when loaded. Defaults
to rgb.
file_client_args (dict): Arguments to instantiate a FileClient.
See :class:`mmcv.fileio.FileClient` for details.
Defaults to dict(backend='disk').
@ -32,11 +34,13 @@ class BaseDataSource(object, metaclass=ABCMeta):
ann_file=None,
test_mode=False,
color_type='color',
channel_order='rgb',
file_client_args=dict(backend='disk')):
self.data_prefix = data_prefix
self.ann_file = ann_file
self.test_mode = test_mode
self.color_type = color_type
self.channel_order = channel_order
self.file_client_args = file_client_args
self.file_client = None
self.CLASSES = self.get_classes(classes)
@ -91,10 +95,16 @@ class BaseDataSource(object, metaclass=ABCMeta):
else:
filename = self.data_infos[idx]['img_info']['filename']
img_bytes = self.file_client.get(filename)
img = mmcv.imfrombytes(img_bytes, flag=self.color_type)
img = mmcv.imfrombytes(
img_bytes,
flag=self.color_type,
channel_order=self.channel_order)
else:
img = self.data_infos[idx]['img']
img_bytes = self.file_client.get(filename)
img = mmcv.imfrombytes(
img_bytes, flag=self.color_type, channel_order=self.channel_order)
img = img.astype(np.uint8)
return Image.fromarray(img)

7
mmselfsup/datasets/pipelines/__init__.py
View File

@ -1,5 +1,8 @@
# Copyright (c) OpenMMLab. All rights reserved.
from .transforms import (GaussianBlur, Lighting, RandomAppliedTrans,
from .transforms import (GaussianBlur, Lighting, RandomAppliedTrans, RandomAug,
Solarization)
__all__ = ['GaussianBlur', 'Lighting', 'RandomAppliedTrans', 'Solarization']
__all__ = [
'GaussianBlur', 'Lighting', 'RandomAppliedTrans', 'Solarization',
'RandomAug'
]

43
mmselfsup/datasets/pipelines/transforms.py
View File

@ -5,6 +5,7 @@ import numpy as np
import torch
from mmcv.utils import build_from_cfg
from PIL import Image, ImageFilter
from timm.data import create_transform
from torchvision import transforms as _transforms
from ..builder import PIPELINES
@ -16,6 +17,48 @@ for m in inspect.getmembers(_transforms, inspect.isclass):
PIPELINES.register_module(m[1])
@PIPELINES.register_module()
class RandomAug(object):
"""RandAugment data augmentation method based on
`"RandAugment: Practical automated data augmentation
with a reduced search space"
<https://arxiv.org/abs/1909.13719>`_.
This code is borrowed from <https://github.com/pengzhiliang/MAE-pytorch>
"""
def __init__(self,
input_size=None,
color_jitter=None,
auto_augment=None,
interpolation=None,
re_prob=None,
re_mode=None,
re_count=None,
mean=None,
std=None):
self.trans = create_transform(
input_size=input_size,
is_training=True,
color_jitter=color_jitter,
auto_augment=auto_augment,
interpolation=interpolation,
re_prob=re_prob,
re_mode=re_mode,
re_count=re_count,
mean=mean,
std=std,
)
def __call__(self, img):
return self.trans(img)
def __repr__(self) -> str:
repr_str = self.__class__.__name__
return repr_str
@PIPELINES.register_module()
class RandomAppliedTrans(object):
"""Randomly applied transformations.

13
mmselfsup/datasets/utils.py
View File

@ -178,9 +178,16 @@ class PrefetchLoader:
for next_input_dict in self.loader:
with torch.cuda.stream(stream):
data = next_input_dict['img'].cuda(non_blocking=True)
next_input_dict['img'] = data.float().sub_(self.mean).div_(
self.std)
if isinstance(next_input_dict['img'], list):
next_input_dict['img'] = [
data.cuda(non_blocking=True).float().sub_(
self.mean).div_(self.std)
for data in next_input_dict['img']
]
else:
data = next_input_dict['img'].cuda(non_blocking=True)
next_input_dict['img'] = data.float().sub_(self.mean).div_(
self.std)
if not first:
yield input_dict # noqa F821

5
mmselfsup/models/algorithms/__init__.py
View File

@ -4,6 +4,7 @@ from .byol import BYOL
from .classification import Classification
from .deepcluster import DeepCluster
from .densecl import DenseCL
from .mae import MAE
from .moco import MoCo
from .mocov3 import MoCoV3
from .npid import NPID
@ -16,6 +17,6 @@ from .swav import SwAV
__all__ = [
'BaseModel', 'BYOL', 'Classification', 'DeepCluster', 'DenseCL', 'MoCo',
'MoCoV3', 'NPID', 'ODC', 'RelativeLoc', 'RotationPred', 'SimCLR',
'SimSiam', 'SwAV'
'NPID', 'ODC', 'RelativeLoc', 'RotationPred', 'SimCLR', 'SimSiam', 'SwAV',
'MAE', 'MoCoV3'
]

16
mmselfsup/models/algorithms/classification.py
View File

@ -1,4 +1,6 @@
# Copyright (c) OpenMMLab. All rights reserved.
from mmcls.models.utils import Augments
from ..builder import ALGORITHMS, build_backbone, build_head
from ..utils import Sobel
from .base import BaseModel
@ -16,7 +18,12 @@ class Classification(BaseModel):
Defaults to None.
"""
def __init__(self, backbone, with_sobel=False, head=None, init_cfg=None):
def __init__(self,
backbone,
with_sobel=False,
head=None,
train_cfg=None,
init_cfg=None):
super(Classification, self).__init__(init_cfg)
self.with_sobel = with_sobel
if with_sobel:
@ -25,6 +32,11 @@ class Classification(BaseModel):
assert head is not None
self.head = build_head(head)
self.augments = None
if train_cfg is not None:
augments_cfg = train_cfg.get('augments', None)
self.augments = Augments(augments_cfg)
def extract_feat(self, img):
"""Function to extract features from backbone.
@ -52,6 +64,8 @@ class Classification(BaseModel):
Returns:
dict[str, Tensor]: A dictionary of loss components.
"""
if self.augments is not None:
img, label = self.augments(img, label)
x = self.extract_feat(img)
outs = self.head(x)
loss_inputs = (outs, label)

58
mmselfsup/models/algorithms/mae.py 100644
View File

@ -0,0 +1,58 @@
# Copyright (c) OpenMMLab. All rights reserved.
from ..builder import ALGORITHMS, build_backbone, build_head, build_neck
from .base import BaseModel
@ALGORITHMS.register_module()
class MAE(BaseModel):
"""MAE.
Implementation of `Masked Autoencoders Are Scalable Vision Learners
<https://arxiv.org/abs/2111.06377>`_.
Args:
backbone (dict): Config dict for encoder. Defaults to None.
neck (dict): Config dict for encoder. Defaults to None.
head (dict): Config dict for loss functions. Defaults to None.
init_cfg (dict): Config dict for weight initialization.
Defaults to None.
"""
def __init__(self, backbone=None, neck=None, head=None, init_cfg=None):
super(MAE, self).__init__(init_cfg)
assert backbone is not None
self.backbone = build_backbone(backbone)
assert neck is not None
self.neck = build_neck(neck)
self.neck.num_patches = self.backbone.patch_embed.num_patches
assert head is not None
self.head = build_head(head)
def init_weights(self):
super(MAE, self).init_weights()
def extract_feat(self, img):
"""Function to extract features from backbone.
Args:
img (Tensor): Input images of shape (N, C, H, W).
Returns:
tuple[Tensor]: backbone outputs.
"""
return self.backbone(img)
def forward_train(self, img, **kwargs):
"""Forward computation during training.
Args:
img (Tensor): Input images of shape (N, C, H, W).
kwargs: Any keyword arguments to be used to forward.
Returns:
dict[str, Tensor]: A dictionary of loss components.
"""
latent, mask, ids_restore = self.backbone(img)
pred = self.neck(latent, ids_restore)
losses = self.head(img, pred, mask)
return losses

7
mmselfsup/models/backbones/__init__.py
View File

@ -1,6 +1,11 @@
# Copyright (c) OpenMMLab. All rights reserved.
from .mae_pretrain_vit import MAEViT
from .mim_cls_vit import MIMVisionTransformer
from .resnet import ResNet, ResNetV1d
from .resnext import ResNeXt
from .vision_transformer import VisionTransformer
__all__ = ['ResNet', 'ResNetV1d', 'ResNeXt', 'VisionTransformer']
__all__ = [
'ResNet', 'ResNetV1d', 'ResNeXt', 'MAEViT', 'MIMVisionTransformer',
'VisionTransformer'
]

156
mmselfsup/models/backbones/mae_pretrain_vit.py 100644
View File

@ -0,0 +1,156 @@
# Copyright (c) OpenMMLab. All rights reserved.
import torch
from mmcls.models import VisionTransformer
from torch import nn
from ..builder import BACKBONES
from ..utils import build_2d_sincos_position_embedding
@BACKBONES.register_module()
class MAEViT(VisionTransformer):
"""Vision Transformer for MAE pre-training.
A PyTorch implement of: `An Image is Worth 16x16 Words: Transformers
for Image Recognition at Scale <https://arxiv.org/abs/2010.11929>`_
Args:
arch (str | dict): Vision Transformer architecture
Default: 'b'
img_size (int | tuple): Input image size
patch_size (int | tuple): The patch size
out_indices (Sequence | int): Output from which stages.
Defaults to -1, means the last stage.
drop_rate (float): Probability of an element to be zeroed.
Defaults to 0.
drop_path_rate (float): stochastic depth rate. Defaults to 0.
norm_cfg (dict): Config dict for normalization layer.
Defaults to ``dict(type='LN')``.
final_norm (bool): Whether to add a additional layer to normalize
final feature map. Defaults to True.
output_cls_token (bool): Whether output the cls_token. If set True,
`with_cls_token` must be True. Defaults to True.
interpolate_mode (str): Select the interpolate mode for position
embeding vector resize. Defaults to "bicubic".
patch_cfg (dict): Configs of patch embeding. Defaults to an empty dict.
layer_cfgs (Sequence | dict): Configs of each transformer layer in
encoder. Defaults to an empty dict.
mask_ratio (bool): The ratio of total number of patches to be masked.
Defaults to 0.75.
init_cfg (dict, optional): Initialization config dict.
Defaults to None.
"""
def __init__(self,
arch='b',
img_size=224,
patch_size=16,
out_indices=-1,
drop_rate=0,
drop_path_rate=0,
norm_cfg=dict(type='LN', eps=1e-6),
final_norm=True,
output_cls_token=True,
interpolate_mode='bicubic',
patch_cfg=dict(),
layer_cfgs=dict(),
mask_ratio=0.75,
init_cfg=None):
super().__init__(arch, img_size, patch_size, out_indices, drop_rate,
drop_path_rate, norm_cfg, final_norm,
output_cls_token, interpolate_mode, patch_cfg,
layer_cfgs, init_cfg)
self.pos_embed.requires_grad = False
self.mask_ratio = mask_ratio
def init_weights(self):
super(MAEViT, self).init_weights()
if not (isinstance(self.init_cfg, dict)
and self.init_cfg['type'] == 'Pretrained'):
# initialize position embedding in backbone
pos_embed = build_2d_sincos_position_embedding(
int(self.patch_embed.num_patches**.5),
self.pos_embed.shape[-1],
cls_token=True)
self.pos_embed.data.copy_(pos_embed.float())
w = self.patch_embed.projection.weight.data
torch.nn.init.xavier_uniform_(w.view([w.shape[0], -1]))
torch.nn.init.normal_(self.cls_token, std=.02)
self.apply(self._init_weights)
def _init_weights(self, m):
if isinstance(m, nn.Linear):
torch.nn.init.xavier_uniform_(m.weight)
if isinstance(m, nn.Linear) and m.bias is not None:
nn.init.constant_(m.bias, 0)
elif isinstance(m, nn.LayerNorm):
nn.init.constant_(m.bias, 0)
nn.init.constant_(m.weight, 1.0)
def random_masking(self, x, mask_ratio=0.75):
"""Generate the mask for MAE Pre-training.
Args:
x (torch.tensor): Image with data augmentation applied.
mask_ratio (float): The mask ratio of total patches.
Defaults to 0.75.
Returns:
tuple[Tensor, Tensor, Tensor]: masked image, mask and the ids
to restore original image.
- x_masked (Tensor): masked image.
- mask (Tensor): mask used to mask image.
- ids_restore (Tensor): ids to restore original image.
"""
N, L, D = x.shape # batch, length, dim
len_keep = int(L * (1 - mask_ratio))
noise = torch.rand(N, L, device=x.device) # noise in [0, 1]
# sort noise for each sample
ids_shuffle = torch.argsort(
noise, dim=1) # ascend: small is keep, large is remove
ids_restore = torch.argsort(ids_shuffle, dim=1)
# keep the first subset
ids_keep = ids_shuffle[:, :len_keep]
x_masked = torch.gather(
x, dim=1, index=ids_keep.unsqueeze(-1).repeat(1, 1, D))
# generate the binary mask: 0 is keep, 1 is remove
mask = torch.ones([N, L], device=x.device)
mask[:, :len_keep] = 0
# unshuffle to get the binary mask
mask = torch.gather(mask, dim=1, index=ids_restore)
return x_masked, mask, ids_restore
def forward(self, x):
B = x.shape[0]
x = self.patch_embed(x)
# add pos embed w/o cls token
x = x + self.pos_embed[:, 1:, :]
# masking: length -> length * mask_ratio
x, mask, ids_restore = self.random_masking(x, self.mask_ratio)
# append cls token
cls_token = self.cls_token + self.pos_embed[:, :1, :]
cls_tokens = cls_token.expand(B, -1, -1)
x = torch.cat((cls_tokens, x), dim=1)
x = self.drop_after_pos(x)
for i, layer in enumerate(self.layers):
x = layer(x)
if i == len(self.layers) - 1 and self.final_norm:
x = self.norm1(x)
return (x, mask, ids_restore)

103
mmselfsup/models/backbones/mim_cls_vit.py 100644
View File

@ -0,0 +1,103 @@
# Copyright (c) OpenMMLab. All rights reserved.
import torch
from mmcls.models import VisionTransformer
from mmcv.cnn import build_norm_layer
from ..builder import BACKBONES
@BACKBONES.register_module()
class MIMVisionTransformer(VisionTransformer):
"""Vision Transformer for MIM-style model (Mask Image Modeling)
classification (fine-tuning or linear probe).
A PyTorch implement of : `An Image is Worth 16x16 Words: Transformers
for Image Recognition at Scale <https://arxiv.org/abs/2010.11929>`_
Args:
arch (str | dict): Vision Transformer architecture
Default: 'b'
img_size (int | tuple): Input image size
patch_size (int | tuple): The patch size
out_indices (Sequence | int): Output from which stages.
Defaults to -1, means the last stage.
drop_rate (float): Probability of an element to be zeroed.
Defaults to 0.
drop_path_rate (float): stochastic depth rate. Defaults to 0.
norm_cfg (dict): Config dict for normalization layer.
Defaults to ``dict(type='LN')``.
final_norm (bool): Whether to add a additional layer to normalize
final feature map. Defaults to True.
output_cls_token (bool): Whether output the cls_token. If set True,
`with_cls_token` must be True. Defaults to True.
interpolate_mode (str): Select the interpolate mode for position
embeding vector resize. Defaults to "bicubic".
patch_cfg (dict): Configs of patch embeding. Defaults to an empty dict.
layer_cfgs (Sequence | dict): Configs of each transformer layer in
encoder. Defaults to an empty dict.
finetune (bool): Whether or not do fine-tuning. Defaults to True.
init_cfg (dict, optional): Initialization config dict.
Defaults to None.
"""
def __init__(self,
arch='b',
img_size=224,
patch_size=16,
out_indices=-1,
drop_rate=0,
drop_path_rate=0,
norm_cfg=dict(type='LN', eps=1e-6),
final_norm=True,
output_cls_token=True,
interpolate_mode='bicubic',
patch_cfg=dict(),
layer_cfgs=dict(),
finetune=True,
init_cfg=None):
super().__init__(arch, img_size, patch_size, out_indices, drop_rate,
drop_path_rate, norm_cfg, final_norm,
output_cls_token, interpolate_mode, patch_cfg,
layer_cfgs, init_cfg)
self.embed_dims = self.arch_settings['embed_dims']
if not self.final_norm:
_, self.fc_norm = build_norm_layer(
norm_cfg, self.embed_dims, postfix=1)
self.finetune = finetune
if not self.finetune:
self._freeze_stages()
def train(self, mode=True):
super(MIMVisionTransformer, self).train(mode)
if not self.finetune:
self._freeze_stages()
def _freeze_stages(self):
"""Freeze params in backbone when linear probing."""
for _, param in self.named_parameters():
param.requires_grad = False
def forward(self, x):
B = x.shape[0]
x = self.patch_embed(x)
# stole cls_tokens impl from Phil Wang, thanks
cls_tokens = self.cls_token.expand(B, -1, -1)
x = torch.cat((cls_tokens, x), dim=1)
x = x + self.pos_embed
x = self.drop_after_pos(x)
for i, layer in enumerate(self.layers):
x = layer(x)
if i == len(self.layers) - 1 and self.final_norm:
x = self.norm1(x)
if not self.final_norm:
x = x[:, 1:, :].mean(dim=1)
outcome = self.fc_norm(x)
else:
outcome = x[:, 0]
return outcome

4
mmselfsup/models/heads/__init__.py
View File

@ -2,11 +2,13 @@
from .cls_head import ClsHead
from .contrastive_head import ContrastiveHead
from .latent_pred_head import LatentClsHead, LatentPredictHead
from .mae_head import MAEFinetuneHead, MAEPretrainHead
from .mocov3_head import MoCoV3Head
from .multi_cls_head import MultiClsHead
from .swav_head import SwAVHead
__all__ = [
'ContrastiveHead', 'ClsHead', 'LatentPredictHead', 'LatentClsHead',
'MoCoV3Head', 'MultiClsHead', 'SwAVHead'
'MultiClsHead', 'SwAVHead', 'MAEFinetuneHead', 'MAEPretrainHead',
'MoCoV3Head'
]

82
mmselfsup/models/heads/mae_head.py 100644
View File

@ -0,0 +1,82 @@
# Copyright (c) OpenMMLab. All rights reserved.
import torch
from mmcls.models import LabelSmoothLoss
from mmcv.cnn.utils.weight_init import trunc_normal_
from mmcv.runner import BaseModule
from torch import nn
from ..builder import HEADS
@HEADS.register_module()
class MAEPretrainHead(BaseModule):
"""Pre-training head for MAE.
Args:
norm_pix_loss (bool): Whether or not normalize target.
Defaults to False.
patch_size (int): Patch size. Defaults to 16.
"""
def __init__(self, norm_pix=False, patch_size=16):
super(MAEPretrainHead, self).__init__()
self.norm_pix = norm_pix
self.patch_size = patch_size
def patchify(self, imgs):
p = self.patch_size
assert imgs.shape[2] == imgs.shape[3] and imgs.shape[2] % p == 0
h = w = imgs.shape[2] // p
x = imgs.reshape(shape=(imgs.shape[0], 3, h, p, w, p))
x = torch.einsum('nchpwq->nhwpqc', x)
x = x.reshape(shape=(imgs.shape[0], h * w, p**2 * 3))
return x
def forward(self, x, pred, mask):
losses = dict()
target = self.patchify(x)
if self.norm_pix:
mean = target.mean(dim=-1, keepdim=True)
var = target.var(dim=-1, keepdim=True)
target = (target - mean) / (var + 1.e-6)**.5
loss = (pred - target)**2
loss = loss.mean(dim=-1)
loss = (loss * mask).sum() / mask.sum()
losses['loss'] = loss
return losses
@HEADS.register_module()
class MAEFinetuneHead(BaseModule):
"""Fine-tuning head for MAE.
Args:
embed_dim (int): The dim of the feature before the classifier head.
num_classes (int): The total classes. Defaults to 1000.
"""
def __init__(self, embed_dim, num_classes=1000, label_smooth_val=0.1):
super(MAEFinetuneHead, self).__init__()
self.head = nn.Linear(embed_dim, num_classes)
self.criterion = LabelSmoothLoss(label_smooth_val, num_classes)
def init_weights(self):
nn.init.constant_(self.head.bias, 0)
trunc_normal_(self.head.weight, std=2e-5)
def forward(self, x):
""""Get the logits."""
outputs = self.head(x)
return [outputs]
def loss(self, outputs, labels):
"""Compute the loss."""
losses = dict()
losses['loss'] = self.criterion(outputs[0], labels)
return losses

4
mmselfsup/models/necks/__init__.py
View File

@ -2,6 +2,7 @@
from .avgpool2d_neck import AvgPool2dNeck
from .densecl_neck import DenseCLNeck
from .linear_neck import LinearNeck
from .mae_neck import MAEPretrainDecoder
from .mocov2_neck import MoCoV2Neck
from .nonlinear_neck import NonLinearNeck
from .odc_neck import ODCNeck
@ -10,5 +11,6 @@ from .swav_neck import SwAVNeck
__all__ = [
'AvgPool2dNeck', 'DenseCLNeck', 'LinearNeck', 'MoCoV2Neck',
'NonLinearNeck', 'ODCNeck', 'RelativeLocNeck', 'SwAVNeck'
'NonLinearNeck', 'ODCNeck', 'RelativeLocNeck', 'SwAVNeck',
'MAEPretrainDecoder'
]

136
mmselfsup/models/necks/mae_neck.py 100644
View File

@ -0,0 +1,136 @@
# Copyright (c) OpenMMLab. All rights reserved.
import torch
import torch.nn as nn
from mmcls.models.backbones.vision_transformer import TransformerEncoderLayer
from mmcv.cnn import build_norm_layer
from mmcv.runner import BaseModule
from ..builder import NECKS
from ..utils import build_2d_sincos_position_embedding
@NECKS.register_module()
class MAEPretrainDecoder(BaseModule):
"""Decoder for MAE Pre-training.
Args:
num_patches (int): The number of total patches. Defaults to 196.
patch_size (int): Image patch size. Defaults to 16.
in_chans (int): The channel of input image. Defaults to 3.
embed_dim (int): Encoder's embedding dimension. Defaults to 1024.
decoder_embed_dim (int): Decoder's embedding dimension.
Defaults to 512.
decoder_depth (int): The depth of decoder. Defaults to 8.
decoder_num_heads (int): Number of attention heads of decoder.
Defaults to 16.
mlp_ratio (int): Ratio of mlp hidden dim to decoder's embedding dim.
Defaults to 4.
norm_cfg (dict): Normalization layer. Defaults to LayerNorm.
Some of the code is borrowed from
`https://github.com/facebookresearch/mae`.
Example:
>>> from mmselfsup.models import MAEPretrainDecoder
>>> import torch
>>> self = MAEPretrainDecoder()
>>> self.eval()
>>> inputs = torch.rand(1, 50, 1024)
>>> ids_restore = torch.arange(0, 196).unsqueeze(0)
>>> level_outputs = self.forward(inputs, ids_restore)
>>> print(tuple(level_outputs.shape))
(1, 196, 768)
"""
def __init__(self,
num_patches=196,
patch_size=16,
in_chans=3,
embed_dim=1024,
decoder_embed_dim=512,
decoder_depth=8,
decoder_num_heads=16,
mlp_ratio=4.,
norm_cfg=dict(type='LN', eps=1e-6)):
super(MAEPretrainDecoder, self).__init__()
self.num_patches = num_patches
self.decoder_embed = nn.Linear(embed_dim, decoder_embed_dim, bias=True)
self.mask_token = nn.Parameter(torch.zeros(1, 1, decoder_embed_dim))
self.decoder_pos_embed = nn.Parameter(
torch.zeros(1, self.num_patches + 1, decoder_embed_dim),
requires_grad=False)
self.decoder_blocks = nn.ModuleList([
TransformerEncoderLayer(
decoder_embed_dim,
decoder_num_heads,
int(mlp_ratio * decoder_embed_dim),
qkv_bias=True,
norm_cfg=norm_cfg) for _ in range(decoder_depth)
])
self.decoder_norm_name, decoder_norm = build_norm_layer(
norm_cfg, decoder_embed_dim, postfix=1)
self.add_module(self.decoder_norm_name, decoder_norm)
self.decoder_pred = nn.Linear(
decoder_embed_dim, patch_size**2 * in_chans, bias=True)
def init_weights(self):
super(MAEPretrainDecoder, self).init_weights()
# initialize position embedding of MAE decoder
decoder_pos_embed = build_2d_sincos_position_embedding(
int(self.num_patches**.5),
self.decoder_pos_embed.shape[-1],
cls_token=True)
self.decoder_pos_embed.data.copy_(decoder_pos_embed.float())
torch.nn.init.normal_(self.mask_token, std=.02)
self.apply(self._init_weights)
def _init_weights(self, m):
if isinstance(m, nn.Linear):
torch.nn.init.xavier_uniform_(m.weight)
if isinstance(m, nn.Linear) and m.bias is not None:
nn.init.constant_(m.bias, 0)
elif isinstance(m, nn.LayerNorm):
nn.init.constant_(m.bias, 0)
nn.init.constant_(m.weight, 1.0)
@property
def decoder_norm(self):
return getattr(self, self.decoder_norm_name)
def forward(self, x, ids_restore):
# embed tokens
x = self.decoder_embed(x)
# append mask tokens to sequence
mask_tokens = self.mask_token.repeat(
x.shape[0], ids_restore.shape[1] + 1 - x.shape[1], 1)
x_ = torch.cat([x[:, 1:, :], mask_tokens], dim=1)
x_ = torch.gather(
x_,
dim=1,
index=ids_restore.unsqueeze(-1).repeat(1, 1, x.shape[2]))
x = torch.cat([x[:, :1, :], x_], dim=1)
# add pos embed
x = x + self.decoder_pos_embed
# apply Transformer blocks
for blk in self.decoder_blocks:
x = blk(x)
x = self.decoder_norm(x)
# predictor projection
x = self.decoder_pred(x)
# remove cls token
x = x[:, 1:, :]
return x

4
mmselfsup/models/utils/__init__.py
View File

@ -10,6 +10,6 @@ from .sobel import Sobel
__all__ = [
'Accuracy', 'accuracy', 'ExtractProcess', 'GatherLayer', 'MultiPooling',
'MultiPrototypes', 'build_2d_sincos_position_embedding', 'ResLayer',
'Sobel'
'MultiPrototypes', 'ResLayer', 'Sobel',
'build_2d_sincos_position_embedding', 'Mixup'
]

4
mmselfsup/utils/__init__.py
View File

@ -14,6 +14,6 @@ __all__ = [
'AliasMethod', 'batch_shuffle_ddp', 'batch_unshuffle_ddp',
'dist_forward_collect', 'nondist_forward_collect', 'collect_env',
'distributed_sinkhorn', 'Extractor', 'concat_all_gather', 'gather_tensors',
'gather_tensors_batch', 'get_root_logger', 'setup_multi_processes',
'multi_gpu_test', 'single_gpu_test'
'gather_tensors_batch', 'get_root_logger', 'multi_gpu_test',
'single_gpu_test', 'setup_multi_processes'
]

8
mmselfsup/utils/clustering.py
View File

@ -1,9 +1,15 @@
# Copyright (c) 2017-present, Facebook, Inc.
# All rights reserved.
# This file is modified from
# https://github.com/facebookresearch/deepcluster/blob/master/clustering.py
import time
import faiss
try:
import faiss
except ImportError:
faiss = None
import numpy as np
import torch
from scipy.sparse import csr_matrix

3
mmselfsup/utils/distributed_sinkhorn.py
View File

@ -1,3 +1,6 @@
# Copyright (c) 2017-present, Facebook, Inc.
# All rights reserved.
# This file is modified from
# https://github.com/facebookresearch/swav/blob/main/main_swav.py

2
mmselfsup/version.py
View File

@ -1,6 +1,6 @@
# Copyright (c) Open-MMLab. All rights reserved.
__version__ = '0.6.0'
__version__ = '0.7.0'
def parse_version_info(version_str):

1
requirements/optional.txt 100644
View File

@ -0,0 +1 @@
faiss-gpu==1.6.1

3
requirements/runtime.txt
View File

@ -1,4 +1,3 @@
faiss-gpu==1.6.1
future
matplotlib
mmcls
@ -8,5 +7,5 @@ scipy
six
sklearn
tensorboard
terminaltables
timm
tqdm

2
setup.cfg
View File

@ -8,7 +8,7 @@ line_length = 79
multi_line_output = 0
extra_standard_library = setuptools
known_first_party = mmselfsup
known_third_party = PIL,cv2,detectron2,faiss,matplotlib,mmcls,mmcv,mmdet,numpy,packaging,pytest,pytorch_sphinx_theme,scipy,seaborn,six,sklearn,svm_helper,torch,torchvision,tqdm
known_third_party = PIL,detectron2,faiss,matplotlib,mmcls,mmcv,mmdet,numpy,packaging,pytest,pytorch_sphinx_theme,scipy,seaborn,six,sklearn,svm_helper,timm,torch,torchvision,tqdm
no_lines_before = STDLIB,LOCALFOLDER
default_section = THIRDPARTY

11
tests/test_data/test_data_sources/test_common.py
View File

@ -25,12 +25,11 @@ def test_data_sources_override_default(dataset_name):
assert dataset.CLASSES == ['bus', 'car']
# Test setting classes through a file
tmp_file = tempfile.NamedTemporaryFile()
with open(tmp_file.name, 'w') as f:
f.write('bus\ncar\n')
dataset = dataset_class(data_prefix='', classes=tmp_file.name)
tmp_file.close()
with tempfile.TemporaryDirectory() as tmpdir:
path = tmpdir + 'classes.txt'
with open(path, 'w') as f:
f.write('bus\ncar\n')
dataset = dataset_class(data_prefix='', classes=path)
assert dataset.CLASSES == ['bus', 'car']
# Test overriding not a subset

8
tests/test_data/test_data_sources/test_image_list.py
View File

@ -1,3 +1,4 @@
# Copyright (c) OpenMMLab. All rights reserved.
import os.path as osp
import pytest
@ -7,8 +8,9 @@ from mmselfsup.datasets.data_sources import ImageList
def test_image_list():
data_source = dict(
data_prefix=osp.join(osp.dirname(__file__), '../../data'),
ann_file=osp.join(osp.dirname(__file__), '../../data/data_list.txt'),
data_prefix=osp.join(osp.dirname(__file__), '..', '..', 'data'),
ann_file=osp.join(
osp.dirname(__file__), '..', '..', 'data', 'data_list.txt'),
)
dataset = ImageList(**data_source)
@ -16,4 +18,4 @@ def test_image_list():
with pytest.raises(AssertionError):
dataset = ImageList(
data_prefix=osp.join(osp.dirname(__file__), '../../data'), )
data_prefix=osp.join(osp.dirname(__file__), '..', '..', 'data'))

3
tests/test_data/test_data_sources/test_imagenet.py
View File

@ -1,3 +1,4 @@
# Copyright (c) OpenMMLab. All rights reserved.
import os.path as osp
import pytest
@ -6,7 +7,7 @@ from mmselfsup.datasets.data_sources import ImageNet
def test_imagenet():
data_source = dict(data_prefix=osp.join(osp.dirname(__file__), '../../'))
data_source = dict(data_prefix='tests')
dataset = ImageNet(**data_source)
assert len(dataset) == 2

4
tests/test_data/test_datasets/test_builder.py
View File

@ -1,3 +1,5 @@
# Copyright (c) OpenMMLab. All rights reserved.
import os.path as osp
from unittest.mock import ANY
import pytest
@ -10,7 +12,7 @@ DATASET_CONFIG = dict(
data_source=dict(
type='ImageNet',
data_prefix=ANY,
ann_file='tests/data/data_list.txt',
ann_file=osp.join('tests', 'data', 'data_list.txt'),
),
pipeline=[
dict(type='RandomResizedCrop', size=224),

4
tests/test_data/test_datasets/test_dataset_wrapper.py
View File

@ -12,9 +12,9 @@ def construct_toy_dataset():
data = dict(
data_source=dict(
type='ImageNet',
data_prefix=osp.join(osp.dirname(__file__), '../../data'),
data_prefix=osp.join(osp.dirname(__file__), '..', '..', 'data'),
ann_file=osp.join(
osp.dirname(__file__), '../../data/data_list.txt'),
osp.dirname(__file__), '..', '..', 'data', 'data_list.txt'),
),
pipeline=[])
dataset = BaseDataset(**data)

4
tests/test_data/test_datasets/test_deepcluster_dataset.py
View File

@ -22,9 +22,9 @@ def test_deepcluster_dataset():
data = dict(
data_source=dict(
type=data_source,
data_prefix=osp.join(osp.dirname(__file__), '../../data'),
data_prefix=osp.join(osp.dirname(__file__), '..', '..', 'data'),
ann_file=osp.join(
osp.dirname(__file__), '../../data/data_list.txt'),
osp.dirname(__file__), '..', '..', 'data', 'data_list.txt'),
),
pipeline=train_pipeline,
prefetch=prefetch)

8
tests/test_data/test_datasets/test_multiview_dataset.py
View File

@ -22,9 +22,9 @@ def test_multi_views_dataste():
data = dict(
data_source=dict(
type=data_source,
data_prefix=osp.join(osp.dirname(__file__), '../../data'),
data_prefix=osp.join(osp.dirname(__file__), '..', '..', 'data'),
ann_file=osp.join(
osp.dirname(__file__), '../../data/data_list.txt'),
osp.dirname(__file__), '..', '..', 'data', 'data_list.txt'),
),
num_views=[2],
pipelines=[train_pipeline, train_pipeline],
@ -36,9 +36,9 @@ def test_multi_views_dataste():
data = dict(
data_source=dict(
type=data_source,
data_prefix=osp.join(osp.dirname(__file__), '../../data'),
data_prefix=osp.join(osp.dirname(__file__), '..', '..', 'data'),
ann_file=osp.join(
osp.dirname(__file__), '../../data/data_list.txt'),
osp.dirname(__file__), '..', '..', 'data', 'data_list.txt'),
),
num_views=[2, 6],
pipelines=[train_pipeline, train_pipeline],

4
tests/test_data/test_datasets/test_relative_loc_dataset.py
View File

@ -22,9 +22,9 @@ def test_relative_loc_dataset():
data = dict(
data_source=dict(
type=data_source,
data_prefix=osp.join(osp.dirname(__file__), '../../data'),
data_prefix=osp.join(osp.dirname(__file__), '..', '..', 'data'),
ann_file=osp.join(
osp.dirname(__file__), '../../data/data_list.txt'),
osp.dirname(__file__), '..', '..', 'data', 'data_list.txt'),
),
pipeline=train_pipeline,
format_pipeline=format_pipeline)

4
tests/test_data/test_datasets/test_rotation_pred_dataset.py
View File

@ -23,9 +23,9 @@ def test_rotation_pred_dataset():
data = dict(
data_source=dict(
type=data_source,
data_prefix=osp.join(osp.dirname(__file__), '../../data'),
data_prefix=osp.join(osp.dirname(__file__), '..', '..', 'data'),
ann_file=osp.join(
osp.dirname(__file__), '../../data/data_list.txt'),
osp.dirname(__file__), '..', '..', 'data', 'data_list.txt'),
),
pipeline=train_pipeline,
prefetch=prefetch)

6
tests/test_data/test_datasets/test_singleview_dataset.py
View File

@ -19,13 +19,13 @@ if not prefetch:
dict(type='Normalize', **img_norm_cfg)])
def test_one_view_dataset():
def test_single_view_dataset():
data = dict(
data_source=dict(
type=data_source,
data_prefix=osp.join(osp.dirname(__file__), '../../data'),
data_prefix=osp.join(osp.dirname(__file__), '..', '..', 'data'),
ann_file=osp.join(
osp.dirname(__file__), '../../data/data_list.txt'),
osp.dirname(__file__), '..', '..', 'data', 'data_list.txt'),
),
pipeline=train_pipeline,
prefetch=prefetch)

35
tests/test_data/test_pipelines.py → tests/test_data/test_pipeline.py
View File

@ -1,6 +1,4 @@
# Copyright (c) OpenMMLab. All rights reserved.
import os.path as osp
import numpy as np
import pytest
import torch
@ -11,7 +9,7 @@ from mmselfsup.datasets.builder import PIPELINES
def test_random_applied_trans():
img = Image.open(osp.join(osp.dirname(__file__), '../data/color.jpg'))
img = Image.fromarray(np.ones((224, 224, 3), dtype=np.uint8))
# p=0.5
transform = dict(
@ -41,8 +39,8 @@ def test_random_applied_trans():
def test_lighting():
transform = dict(type='Lighting')
module = build_from_cfg(transform, PIPELINES)
img = np.array(
Image.open(osp.join(osp.dirname(__file__), '../data/color.jpg')))
img = np.ones((224, 224, 3), dtype=np.uint8)
with pytest.raises(AssertionError):
res = module(img)
@ -58,7 +56,7 @@ def test_gaussianblur():
type='GaussianBlur', sigma_min=0.1, sigma_max=1.0, p=-1)
module = build_from_cfg(transform, PIPELINES)
img = Image.open(osp.join(osp.dirname(__file__), '../data/color.jpg'))
img = Image.fromarray(np.ones((224, 224, 3), dtype=np.uint8))
# p=0.5
transform = dict(type='GaussianBlur', sigma_min=0.1, sigma_max=1.0)
@ -81,7 +79,7 @@ def test_solarization():
transform = dict(type='Solarization', p=-1)
module = build_from_cfg(transform, PIPELINES)
img = Image.open(osp.join(osp.dirname(__file__), '../data/color.jpg'))
img = Image.fromarray(np.ones((224, 224, 3), dtype=np.uint8))
# p=0.5
transform = dict(type='Solarization')
@ -97,3 +95,26 @@ def test_solarization():
res = module(img)
assert img.size == res.size
def test_randomaug():
transform = dict(
type='RandomAug',
input_size=224,
color_jitter=None,
auto_augment='rand-m9-mstd0.5-inc1',
interpolation='bicubic',
re_prob=0.25,
re_mode='pixel',
re_count=1,
mean=(0.485, 0.456, 0.406),
std=(0.229, 0.224, 0.225))
img = Image.fromarray(np.uint8(np.ones((224, 224, 3))))
module = build_from_cfg(transform, PIPELINES)
res = module(img)
assert list(res.shape) == [3, 224, 224]
assert isinstance(str(module), str)

7
tests/test_data/test_utils.py
View File

@ -1,17 +1,20 @@
# Copyright (c) OpenMMLab. All rights reserved.
import os.path as osp
import platform
import random
import string
import tempfile
import numpy as np
import pytest
from PIL import Image
from mmselfsup.datasets.utils import check_integrity, rm_suffix, to_numpy
def test_to_numpy():
pil_img = Image.open(osp.join(osp.dirname(__file__), '../data/color.jpg'))
pil_img = Image.open(
osp.join(osp.dirname(__file__), '..', 'data', 'color.jpg'))
np_img = to_numpy(pil_img)
assert type(np_img) == np.ndarray
if np_img.ndim < 3:
@ -20,6 +23,8 @@ def test_to_numpy():
assert np_img.shape[0] == 3
@pytest.mark.skipif(
platform.system() == 'Windows', reason='Windows permission')
def test_dataset_utils():
# test rm_suffix
assert rm_suffix('a.jpg') == 'a'

3
tests/test_models/test_algorithms/test_byol.py
View File

@ -1,4 +1,6 @@
# Copyright (c) OpenMMLab. All rights reserved.
import platform
import pytest
import torch
@ -32,6 +34,7 @@ head = dict(
norm_cfg=dict(type='BN1d')))
@pytest.mark.skipif(platform.system() == 'Windows', reason='Windows mem limit')
def test_byol():
with pytest.raises(AssertionError):
alg = BYOL(backbone=backbone, neck=None, head=head)

4
tests/test_models/test_algorithms/test_classification.py
View File

@ -1,9 +1,13 @@
# Copyright (c) OpenMMLab. All rights reserved.
import platform
import pytest
import torch
from mmselfsup.models.algorithms import Classification
@pytest.mark.skipif(platform.system() == 'Windows', reason='Windows mem limit')
def test_classification():
# test ResNet
with_sobel = True,

3
tests/test_models/test_algorithms/test_deepcluster.py
View File

@ -1,4 +1,6 @@
# Copyright (c) OpenMMLab. All rights reserved.
import platform
import pytest
import torch
@ -20,6 +22,7 @@ head = dict(
num_classes=num_classes)
@pytest.mark.skipif(platform.system() == 'Windows', reason='Windows mem limit')
def test_deepcluster():
with pytest.raises(AssertionError):
alg = DeepCluster(

3
tests/test_models/test_algorithms/test_densecl.py
View File

@ -1,4 +1,6 @@
# Copyright (c) OpenMMLab. All rights reserved.
import platform
import pytest
import torch
@ -23,6 +25,7 @@ neck = dict(
head = dict(type='ContrastiveHead', temperature=0.2)
@pytest.mark.skipif(platform.system() == 'Windows', reason='Windows mem limit')
def test_densecl():
with pytest.raises(AssertionError):
alg = DenseCL(backbone=backbone, neck=None, head=head)

37
tests/test_models/test_algorithms/test_mae.py 100644
View File

@ -0,0 +1,37 @@
# Copyright (c) OpenMMLab. All rights reserved.
import platform
import pytest
import torch
from mmselfsup.models.algorithms import MAE
backbone = dict(type='MAEViT', arch='b', patch_size=16, mask_ratio=0.75)
neck = dict(
type='MAEPretrainDecoder',
patch_size=16,
in_chans=3,
embed_dim=768,
decoder_embed_dim=512,
decoder_depth=8,
decoder_num_heads=16,
mlp_ratio=4.,
)
head = dict(type='MAEPretrainHead', norm_pix=False, patch_size=16)
@pytest.mark.skipif(platform.system() == 'Windows', reason='Windows mem limit')
def test_mae():
with pytest.raises(AssertionError):
alg = MAE(backbone=backbone, neck=None, head=head)
with pytest.raises(AssertionError):
alg = MAE(backbone=backbone, neck=neck, head=None)
with pytest.raises(AssertionError):
alg = MAE(backbone=None, neck=neck, head=head)
alg = MAE(backbone=backbone, neck=neck, head=head)
fake_input = torch.randn((16, 3, 224, 224))
fake_loss = alg.forward_train(fake_input)
fake_feature = alg.extract_feat(fake_input)
assert isinstance(fake_loss['loss'].item(), float)
assert list(fake_feature[0].shape) == [16, 50, 768]

3
tests/test_models/test_algorithms/test_moco.py
View File

@ -1,4 +1,6 @@
# Copyright (c) OpenMMLab. All rights reserved.
import platform
import pytest
import torch
@ -22,6 +24,7 @@ neck = dict(
head = dict(type='ContrastiveHead', temperature=0.2)
@pytest.mark.skipif(platform.system() == 'Windows', reason='Windows mem limit')
def test_moco():
with pytest.raises(AssertionError):
alg = MoCo(backbone=backbone, neck=None, head=head)

3
tests/test_models/test_algorithms/test_mocov3.py
View File

@ -1,4 +1,6 @@
# Copyright (c) OpenMMLab. All rights reserved.
import platform
import pytest
import torch
@ -38,6 +40,7 @@ head = dict(
temperature=0.2)
@pytest.mark.skipif(platform.system() == 'Windows', reason='Windows mem limit')
def test_mocov3():
with pytest.raises(AssertionError):
alg = MoCoV3(backbone=backbone, neck=None, head=head)

5
tests/test_models/test_algorithms/test_npid.py
View File

@ -1,4 +1,6 @@
# Copyright (c) OpenMMLab. All rights reserved.
import platform
import pytest
import torch
@ -17,7 +19,8 @@ memory_bank = dict(type='SimpleMemory', length=8, feat_dim=4, momentum=0.5)
@pytest.mark.skipif(
not torch.cuda.is_available(), reason='CUDA is not available.')
not torch.cuda.is_available() or platform.system() == 'Windows',
reason='CUDA is not available or Windows mem limit')
def test_npid():
with pytest.raises(AssertionError):
alg = NPID(backbone=backbone, neck=neck, head=head, memory_bank=None)

5
tests/test_models/test_algorithms/test_odc.py
View File

@ -1,4 +1,6 @@
# Copyright (c) OpenMMLab. All rights reserved.
import platform
import pytest
import torch
@ -33,7 +35,8 @@ memory_bank = dict(
@pytest.mark.skipif(
not torch.cuda.is_available(), reason='CUDA is not available.')
not torch.cuda.is_available() or platform.system() == 'Windows',
reason='CUDA is not available or Windows mem limit')
def test_odc():
with pytest.raises(AssertionError):
alg = ODC(backbone=backbone, neck=neck, head=head, memory_bank=None)

3
tests/test_models/test_algorithms/test_relative_loc.py
View File

@ -1,4 +1,6 @@
# Copyright (c) OpenMMLab. All rights reserved.
import platform
import pytest
import torch
@ -18,6 +20,7 @@ neck = dict(
head = dict(type='ClsHead', with_avg_pool=False, in_channels=4, num_classes=8)
@pytest.mark.skipif(platform.system() == 'Windows', reason='Windows mem limit')
def test_relative_loc():
with pytest.raises(AssertionError):
alg = RelativeLoc(backbone=backbone, neck=None, head=head)

3
tests/test_models/test_algorithms/test_rotation_pred.py
View File

@ -1,4 +1,6 @@
# Copyright (c) OpenMMLab. All rights reserved.
import platform
import pytest
import torch
@ -14,6 +16,7 @@ head = dict(
type='ClsHead', with_avg_pool=True, in_channels=2048, num_classes=4)
@pytest.mark.skipif(platform.system() == 'Windows', reason='Windows mem limit')
def test_rotation_pred():
with pytest.raises(AssertionError):
alg = RotationPred(backbone=backbone, head=None)

3
tests/test_models/test_algorithms/test_simclr.py
View File

@ -1,4 +1,6 @@
# Copyright (c) OpenMMLab. All rights reserved.
import platform
import pytest
import torch
@ -20,6 +22,7 @@ neck = dict(
head = dict(type='ContrastiveHead', temperature=0.1)
@pytest.mark.skipif(platform.system() == 'Windows', reason='Windows mem limit')
def test_simclr():
with pytest.raises(AssertionError):
alg = SimCLR(backbone=backbone, neck=None, head=head)

3
tests/test_models/test_algorithms/test_simsiam.py
View File

@ -1,4 +1,6 @@
# Copyright (c) OpenMMLab. All rights reserved.
import platform
import pytest
import torch
@ -33,6 +35,7 @@ head = dict(
norm_cfg=dict(type='BN1d')))
@pytest.mark.skipif(platform.system() == 'Windows', reason='Windows mem limit')
def test_simsiam():
with pytest.raises(AssertionError):
alg = SimSiam(backbone=backbone, neck=neck, head=None)

Some files were not shown because too many files have changed in this diff Show More