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

Reformat config files (#46) 

* fix init

* fix test api

fix test api bug

* add metarcnn fsdetview config

* update pr

* update meta file

* fix CI
pull/1/head
Linyiqi 2021-11-01 23:27:01 +08:00 committed by GitHub
parent 37cd8a1f7d
commit 3508afca93
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
459 changed files with 8904 additions and 2416 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
.github/workflows/build.yml vendored
View File

@ -93,11 +93,6 @@ jobs:
torch_version: torch1.6.0
torchvision: 0.7.0+cu101
mmcv: "latest+torch1.6.0+cu101"
- torch: 1.6.0+cu101
torch_version: torch1.6.0
torchvision: 0.7.0+cu101
mmcv: "latest+torch1.6.0+cu101"
python-version: 3.6
- torch: 1.6.0+cu101
torch_version: torch1.6.0
torchvision: 0.7.0+cu101

5
.pre-commit-config.yaml
View File

@ -38,3 +38,8 @@ repos:
hooks:
- id: docformatter
args: ["--in-place", "--wrap-descriptions", "79"]
- repo: https://github.com/codespell-project/codespell
rev: v2.1.0
hooks:
- id: codespell
args: [ "--skip", "*.ipynb", "-L", "confectionary"]

122
README.md
View File

@ -0,0 +1,122 @@
[comment]: <> (<div align="center">)
[comment]: <> ( <img src="resources/mmfewshot-logo.png" width="500px"/>)
[comment]: <> (</div>)
## Introduction
[comment]: <> (English | [简体中文]&#40;/README_zh-CN.md&#41;)
[comment]: <> ([![Documentation]&#40;https://readthedocs.org/projects/mmfewshot/badge/?version=latest&#41;]&#40;https://mmfewshot.readthedocs.io/en/latest/?badge=latest&#41;)
[comment]: <> ([![actions]&#40;https://github.com/open-mmlab/mmfewshot/workflows/build/badge.svg&#41;]&#40;https://github.com/open-mmlab/mmfewshot/actions&#41;)
[comment]: <> ([![codecov]&#40;https://codecov.io/gh/open-mmlab/mmfewshot/branch/master/graph/badge.svg&#41;]&#40;https://codecov.io/gh/open-mmlab/mmfewshot&#41;)
[comment]: <> ([![PyPI]&#40;https://badge.fury.io/py/mmedit.svg&#41;]&#40;https://pypi.org/project/mmedit/&#41;)
[comment]: <> ([![LICENSE]&#40;https://img.shields.io/github/license/open-mmlab/mmfewshot.svg&#41;]&#40;https://github.com/open-mmlab/mmfewshot/blob/master/LICENSE&#41;)
[comment]: <> ([![Average time to resolve an issue]&#40;https://isitmaintained.com/badge/resolution/open-mmlab/mmfewshot.svg&#41;]&#40;https://github.com/open-mmlab/mmfewshot/issues&#41;)
[comment]: <> ([![Percentage of issues still open]&#40;https://isitmaintained.com/badge/open/open-mmlab/mmfewshot.svg&#41;]&#40;https://github.com/open-mmlab/mmfewshot/issues&#41;)
mmfewshot is an open source few shot learning toolbox based on PyTorch. It is a part of the [OpenMMLab](https://open-mmlab.github.io/) project.
The master branch works with **PyTorch 1.7+**.
The compatibility to earlier versions of PyTorch is not fully tested.
Documentation: https://mmfewshot.readthedocs.io/en/latest/.
### Major features
## Model Zoo
Supported algorithms:
<details open>
<summary>classification</summary>
- [x] [Baseline](configs/classification/baseline/README.md) (ICLR'2019)
- [x] [Baseline++](configs/classification/baseline_plus/README.md) (ICLR'2019)
- [x] [NegMargin](configs/classification/neg_margin/README.md) (ECCV'2020)
- [x] [MatchingNet](configs/classification/matching_net/README.md) (NeurIPS'2016)
- [x] [ProtoNet](configs/classification/proto_net/README.md) (NeurIPS'2017)
- [x] [RelationNet](configs/classification/relation_net/README.md) (CVPR'2018)
- [x] [MetaBaseline](configs/classification/meta_baseline/README.md) (ICCV'2021)
- [x] [MAML](configs/classification/maml/README.md) (ICML'2017)
</details>
<details open>
<summary>Detection</summary>
- [x] [TFA](configs/detection/tfa/README.md) (ICML'2020)
- [x] [FSCE](configs/detection/fsce/README.md) (CVPR'2021)
- [x] [AttentionRPN](configs/detection/attention_rpn/README.md) (CVPR'2020)
- [x] [MetaRCNN](configs/detection/meta_rcnn/README.md) (ICCV'2019)
- [x] [FSDetView](configs/detection/fsdetview/README.md) (ECCV'2020)
- [x] [MPSR](configs/detection/mpsr/README.md) (ECCV'2020)
</details>
Please refer to [model_zoo](https://mmfewshot.readthedocs.io/en/latest/modelzoo.html) for more details.
## License
This project is released under the [Apache 2.0 license](LICENSE).
## Changelog
## Installation
Please refer to [install.md](docs/get_started.md) for installation.
## Get Started
Please see [getting_started.md](docs/get_started.md) for the basic usage of mmfewshot.
## Citation
If you find this project useful in your research, please consider cite:
```bibtex
@misc{mmfewshot2020,
title={OpenMMLab Few Shot Learning Toolbox and Benchmark},
author={mmfewshot Contributors},
howpublished = {\url{https://github.com/open-mmlab/mmfewshot}},
year={2021}
}
```
## Contributing
We appreciate all contributions to improve mmfewshot. Please refer to [CONTRIBUTING.md in MMFewShot](https://github.com/open-mmlab/mmcv/blob/master/.github/CONTRIBUTING.md) for the contributing guideline.
## Acknowledgement
mmfewshot is an open source project that is contributed by researchers and engineers from various colleges and companies. We appreciate all the contributors who implement their methods or add new features, as well as users who give valuable feedbacks. We wish that the toolbox and benchmark could serve the growing research community by providing a flexible toolkit to reimplement existing methods and develop their own new methods.
## 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.
- [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.
- [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.
- [MMPose](https://github.com/open-mmlab/mmpose): OpenMMLab pose estimation toolbox and benchmark.
- [MMEditing](https://github.com/open-mmlab/mmfewshot): OpenMMLab image and video editing toolbox.
- [MMOCR](https://github.com/open-mmlab/mmocr): A Comprehensive Toolbox for Text Detection, Recognition and Understanding.
- [MMGeneration](https://github.com/open-mmlab/mmgeneration): A powerful toolkit for generative models.

11
configs/classification/README.md 100644
View File

@ -0,0 +1,11 @@
# Few Shot Classification
## Get Started
To be done.
## Data Preparation
To be done.
## Model Zoo
To be done.

6
configs/classification/_base_/meta_test/cub_meta_test_5way_1shot_84x84.py → configs/classification/_base_/meta_test/cub_meta-test_5way-1shot.py
View File

@ -31,7 +31,7 @@ data = dict(
# testing acceleration.
fast_test=True,
test_set=dict(batch_size=16, num_workers=2),
# initialize worker is a time consuming operation
# worker initialization is a time consuming operation
support=dict(batch_size=num_ways * num_shots, num_workers=0),
query=dict(batch_size=num_ways * num_queries, num_workers=0))),
test=dict(
@ -42,7 +42,7 @@ data = dict(
data_prefix='data/CUB_200_2011',
pipeline=test_pipeline),
meta_test_cfg=dict(
num_episodes=1000,
num_episodes=2000,
num_ways=num_ways,
num_shots=num_shots,
num_queries=num_queries,
@ -50,6 +50,6 @@ data = dict(
# testing acceleration.
fast_test=True,
test_set=dict(batch_size=16, num_workers=2),
# initialize worker is a time consuming operation
# worker initialization is a time consuming operation
support=dict(batch_size=num_ways * num_shots, num_workers=0),
query=dict(batch_size=num_ways * num_queries, num_workers=0))))

6
configs/classification/_base_/meta_test/cub_meta_test_5way_5shot_84x84.py → configs/classification/_base_/meta_test/cub_meta-test_5way-5shot.py
View File

@ -31,7 +31,7 @@ data = dict(
# testing acceleration.
fast_test=True,
test_set=dict(batch_size=16, num_workers=2),
# initialize worker is a time consuming operation
# worker initialization is a time consuming operation
support=dict(batch_size=num_ways * num_shots, num_workers=0),
query=dict(batch_size=num_ways * num_queries, num_workers=0))),
test=dict(
@ -42,7 +42,7 @@ data = dict(
data_prefix='data/CUB_200_2011',
pipeline=test_pipeline),
meta_test_cfg=dict(
num_episodes=1000,
num_episodes=2000,
num_ways=num_ways,
num_shots=num_shots,
num_queries=num_queries,
@ -50,6 +50,6 @@ data = dict(
# testing acceleration.
fast_test=True,
test_set=dict(batch_size=16, num_workers=2),
# initialize worker is a time consuming operation
# worker initialization is a time consuming operation
support=dict(batch_size=num_ways * num_shots, num_workers=0),
query=dict(batch_size=num_ways * num_queries, num_workers=0))))

6
configs/classification/_base_/meta_test/mini_imagenet_meta_test_5way_1shot_84x84.py → configs/classification/_base_/meta_test/mini-imagenet_meta-test_5way-1shot.py
View File

@ -31,7 +31,7 @@ data = dict(
# testing acceleration.
fast_test=True,
test_set=dict(batch_size=16, num_workers=2),
# initialize worker is a time consuming operation
# worker initialization is a time consuming operation
support=dict(batch_size=num_ways * num_shots, num_workers=0),
query=dict(batch_size=num_ways * num_queries, num_workers=0))),
test=dict(
@ -42,7 +42,7 @@ data = dict(
data_prefix='data/mini_imagenet',
pipeline=test_pipeline),
meta_test_cfg=dict(
num_episodes=1000,
num_episodes=2000,
num_ways=num_ways,
num_shots=num_shots,
num_queries=num_queries,
@ -50,6 +50,6 @@ data = dict(
# testing acceleration.
fast_test=True,
test_set=dict(batch_size=16, num_workers=2),
# initialize worker is a time consuming operation
# worker initialization is a time consuming operation
support=dict(batch_size=num_ways * num_shots, num_workers=0),
query=dict(batch_size=num_ways * num_queries, num_workers=0))))

6
configs/classification/_base_/meta_test/mini_imagenet_meta_test_5way_5shot_84x84.py → configs/classification/_base_/meta_test/mini-imagenet_meta-test_5way-5shot.py
View File

@ -31,7 +31,7 @@ data = dict(
# testing acceleration.
fast_test=True,
test_set=dict(batch_size=16, num_workers=2),
# initialize worker is a time consuming operation
# worker initialization is a time consuming operation
support=dict(batch_size=num_ways * num_shots, num_workers=0),
query=dict(batch_size=num_ways * num_queries, num_workers=0))),
test=dict(
@ -42,7 +42,7 @@ data = dict(
data_prefix='data/mini_imagenet',
pipeline=test_pipeline),
meta_test_cfg=dict(
num_episodes=1000,
num_episodes=2000,
num_ways=num_ways,
num_shots=num_shots,
num_queries=num_queries,
@ -50,6 +50,6 @@ data = dict(
# testing acceleration.
fast_test=True,
test_set=dict(batch_size=16, num_workers=2),
# initialize worker is a time consuming operation
# worker initialization is a time consuming operation
support=dict(batch_size=num_ways * num_shots, num_workers=0),
query=dict(batch_size=num_ways * num_queries, num_workers=0))))

6
configs/classification/_base_/meta_test/tiered_imagenet_meta_test_5way_1shot_84x84.py → configs/classification/_base_/meta_test/tiered-imagenet_meta-test_5way-1shot.py
View File

@ -31,7 +31,7 @@ data = dict(
# testing acceleration.
fast_test=True,
test_set=dict(batch_size=16, num_workers=2),
# initialize worker is a time consuming operation
# worker initialization is a time consuming operation
support=dict(batch_size=num_ways * num_shots, num_workers=0),
query=dict(batch_size=num_ways * num_queries, num_workers=0))),
test=dict(
@ -42,7 +42,7 @@ data = dict(
data_prefix='data/tiered_imagenet',
pipeline=test_pipeline),
meta_test_cfg=dict(
num_episodes=1000,
num_episodes=2000,
num_ways=num_ways,
num_shots=num_shots,
num_queries=num_queries,
@ -50,6 +50,6 @@ data = dict(
# testing acceleration.
fast_test=True,
test_set=dict(batch_size=16, num_workers=2),
# initialize worker is a time consuming operation
# worker initialization is a time consuming operation
support=dict(batch_size=num_ways * num_shots, num_workers=0),
query=dict(batch_size=num_ways * num_queries, num_workers=0))))

6
configs/classification/_base_/meta_test/tiered_imagenet_meta_test_5way_5shot_84x84.py → configs/classification/_base_/meta_test/tiered-imagenet_meta-test_5way-5shot.py
View File

@ -31,7 +31,7 @@ data = dict(
# testing acceleration.
fast_test=True,
test_set=dict(batch_size=16, num_workers=2),
# initialize worker is a time consuming operation
# worker initialization is a time consuming operation
support=dict(batch_size=num_ways * num_shots, num_workers=0),
query=dict(batch_size=num_ways * num_queries, num_workers=0))),
test=dict(
@ -42,7 +42,7 @@ data = dict(
data_prefix='data/tiered_imagenet',
pipeline=test_pipeline),
meta_test_cfg=dict(
num_episodes=1000,
num_episodes=2000,
num_ways=num_ways,
num_shots=num_shots,
num_queries=num_queries,
@ -50,6 +50,6 @@ data = dict(
# testing acceleration.
fast_test=True,
test_set=dict(batch_size=16, num_workers=2),
# initialize worker is a time consuming operation
# worker initialization is a time consuming operation
support=dict(batch_size=num_ways * num_shots, num_workers=0),
query=dict(batch_size=num_ways * num_queries, num_workers=0))))

2
configs/classification/_base_/runtime/epoch_based_runtime.py
View File

@ -3,7 +3,6 @@ log_config = dict(
interval=50,
hooks=[
dict(type='TextLoggerHook'),
# dict(type='TensorboardLoggerHook')
])
# yapf:enable
checkpoint_config = dict(interval=50)
@ -15,3 +14,4 @@ resume_from = None
workflow = [('train', 1)]
pin_memory = True
use_infinite_sampler = True
seed = 0

2
configs/classification/_base_/runtime/iter_based_runtime.py
View File

@ -3,7 +3,6 @@ log_config = dict(
interval=50,
hooks=[
dict(type='TextLoggerHook'),
# dict(type='TensorboardLoggerHook')
])
# yapf:enable
checkpoint_config = dict(interval=4000)
@ -15,3 +14,4 @@ resume_from = None
workflow = [('train', 1)]
pin_memory = True
use_infinite_sampler = True
seed = 0

0
configs/classification/_base_/schedules/adam_100000iter.py → configs/classification/_base_/schedules/adam_100k_iter.py
View File

4
configs/classification/_base_/schedules/adam_200epoch.py
View File

@ -1,4 +0,0 @@
runner = dict(type='EpochBasedRunner', max_epochs=200)
optimizer = dict(type='Adam', lr=0.001)
optimizer_config = dict(grad_clip=None)
lr_config = dict(policy='fixed', warmup=None)

2
configs/classification/_base_/schedules/sgd_60000iter.py → configs/classification/_base_/schedules/sgd_100k_iter.py
View File

@ -1,4 +1,4 @@
runner = dict(type='IterBasedRunner', max_iters=60000)
runner = dict(type='IterBasedRunner', max_iters=100000)
optimizer = dict(type='SGD', lr=0.001, momentum=0.9, weight_decay=0.0001)
optimizer_config = dict(grad_clip=None)
# learning policy

69
configs/classification/baseline/README.md 100644
View File

@ -0,0 +1,69 @@
<!-- [ALGORITHM] -->
# <summary><a href="https://arxiv.org/abs/1904.04232"> Baseline (ICLR'2019)</a></summary>
```bibtex
@inproceedings{chen2019closerfewshot,
title={A Closer Look at Few-shot Classification},
author={Chen, Wei-Yu and Liu, Yen-Cheng and Kira, Zsolt and Wang, Yu-Chiang and Huang, Jia-Bin},
booktitle={International Conference on Learning Representations},
year={2019}
}
```
## How to Reproduce Baseline
It consists of two steps:
- **Step1: Base training**
- use all the images of base classes to train a base model.
- use validation set to select the best model.
- **Step2: Meta Testing**:
- use best model from step1.
### An example of CUB dataset with Conv4
```bash
# base training
python ./tools/classification/train.py \
configs/classification/baseline/cub/baseline_conv4_1xb64_cub_5way-1shot.py
# meta testing
python ./tools/classification/test.py \
configs/classification/baseline/cub/baseline_conv4_1xb64_cub_5way-1shot.py \
work_dir/baseline_conv4_1xb64_cub_5way-1shot/best_accuracy_mean.pth
```
**Note**:
- All the result are trained with single gpu.
- The base training of 1 shot and 5 shot use same training setting,
but different validation setting.
## Results on CUB dataset of 2000 episodes
| Arch | Input Size | Batch Size | way | shot | mean Acc | std | ckpt | log |
| :-------------- | :-----------: | :------: | :------: | :------: | :------: | :------: |:------: |:------: |
| [conv4](/configs/classification/baseline/cub/baseline_conv4_1xb64_cub_5way-1shot.py) | 84x84 | 64 | 5 | 1 | - | - | [ckpt]() | [log]() |
| [conv4](/configs/classification/baseline/cub/baseline_conv4_1xb64_cub_5way-5shot.py) | 84x84 | 64 | 5 | 5 | - | - | [ckpt]() | [log]() |
| [resnet12](/configs/classification/baseline/cub/baseline_resnet12_1xb64_cub_5way-1shot.py) | 84x84 | 64 | 5 | 1 | - | - | [ckpt]() | [log]() |
| [resnet12](/configs/classification/baseline/cub/baseline_resnet12_1xb64_cub_5way-5shot.py) | 84x84 | 64 | 5 | 5 | - | - | [ckpt]() | [log]() |
## Results on Mini-ImageNet dataset of 2000 episodes
| Arch | Input Size | Batch Size | way | shot | mean Acc | std | ckpt | log |
| :-------------- | :-----------: | :------: | :------: | :------: | :------: | :------: |:------: |:------: |
| [conv4](/configs/classification/baseline/mini_imagenet/baseline_conv4_1xb64_mini-imagenet_5way-1shot.py) | 84x84 | 64 | 5 | 1 | - | - | [ckpt]() | [log]() |
| [conv4](/configs/classification/baseline/mini_imagenet/baseline_conv4_1xb64_mini-imagenet_5way-5shot.py) | 84x84 | 64 | 5 | 5 | - | - | [ckpt]() | [log]() |
| [resnet12](/configs/classification/baseline/mini_imagenet/baseline_resnet12_1xb64_mini-imagenet_5way-1shot.py) | 84x84 | 64 | 5 | 1 | - | - | [ckpt]() | [log]() |
| [resnet12](/configs/classification/baseline/mini_imagenet/baseline_resnet12_1xb64_mini-imagenet_5way-5shot.py) | 84x84 | 64 | 5 | 5 | - | - | [ckpt]() | [log]() |
## Results on Tiered-ImageNet of 2000 episodes
| Arch | Input Size | Batch Size | way | shot | mean Acc | std | ckpt | log |
| :-------------- | :-----------: | :------: | :------: | :------: | :------: | :------: |:------: |:------: |
| [conv4](/configs/classification/baseline/tiered_imagenet/baseline_conv4_1xb64_tiered-imagenet_5way-1shot.py) | 84x84 | 64 | 5 | 1 | - | - | [ckpt]() | [log]() |
| [conv4](/configs/classification/baseline/tiered_imagenet/baseline_conv4_1xb64_tiered-imagenet_5way-5shot.py) | 84x84 | 64 | 5 | 5 | - | - | [ckpt]() | [log]() |
| [resnet12](/configs/classification/baseline/tiered_imagenet/baseline_resnet12_1xb64_tiered-imagenet_5way-1shot.py) | 84x84 | 64 | 5 | 1 | - | - | [ckpt]() | [log]() |
| [resnet12](/configs/classification/baseline/tiered_imagenet/baseline_resnet12_1xb64_tiered-imagenet_5way-5shot.py) | 84x84 | 64 | 5 | 5 | - | - | [ckpt]() | [log]() |

10
configs/classification/baseline_plus/cub/baseline_plus_cub_5way_1shot_84x84_aug.py → configs/classification/baseline/cub/baseline_conv4_1xb64_cub_5way-1shot.py
View File

@ -1,5 +1,5 @@
_base_ = [
'../../_base_/meta_test/cub_meta_test_5way_1shot_84x84.py',
'../../_base_/meta_test/cub_meta-test_5way-1shot.py',
'../../_base_/runtime/epoch_based_runtime.py',
'../../_base_/schedules/sgd_200epoch.py'
]
@ -25,7 +25,7 @@ meta_finetune_cfg = dict(
data = dict(
samples_per_gpu=64,
workers_per_gpu=6,
workers_per_gpu=4,
train=dict(
type='CUBDataset',
data_prefix='data/CUB_200_2011',
@ -39,3 +39,9 @@ data = dict(
meta_test_cfg=dict(
support=dict(
batch_size=4, drop_last=True, train=meta_finetune_cfg))))
model = dict(
type='BaselineClassifier',
backbone=dict(type='Conv4'),
head=dict(type='LinearHead', num_classes=100, in_channels=1600),
meta_test_head=dict(type='LinearHead', num_classes=5, in_channels=1600))

10
configs/classification/baseline_plus/cub/baseline_plus_cub_5way_5shot_84x84_aug.py → configs/classification/baseline/cub/baseline_conv4_1xb64_cub_5way-5shot.py
View File

@ -1,5 +1,5 @@
_base_ = [
'../../_base_/meta_test/cub_meta_test_5way_5shot_84x84.py',
'../../_base_/meta_test/cub_meta-test_5way-5shot.py',
'../../_base_/runtime/epoch_based_runtime.py',
'../../_base_/schedules/sgd_200epoch.py'
]
@ -25,7 +25,7 @@ meta_finetune_cfg = dict(
data = dict(
samples_per_gpu=64,
workers_per_gpu=6,
workers_per_gpu=4,
train=dict(
type='CUBDataset',
data_prefix='data/CUB_200_2011',
@ -39,3 +39,9 @@ data = dict(
meta_test_cfg=dict(
support=dict(
batch_size=4, drop_last=True, train=meta_finetune_cfg))))
model = dict(
type='BaselineClassifier',
backbone=dict(type='Conv4'),
head=dict(type='LinearHead', num_classes=100, in_channels=1600),
meta_test_head=dict(type='LinearHead', num_classes=5, in_channels=1600))

7
configs/classification/baseline/cub/baseline_conv4_cub_5way_1shot.py
View File

@ -1,7 +0,0 @@
_base_ = ['baseline_cub_5way_1shot_84x84_aug.py']
model = dict(
type='BaselineClassifier',
backbone=dict(type='Conv4'),
head=dict(type='LinearHead', num_classes=100, in_channels=1600),
meta_test_head=dict(type='LinearHead', num_classes=5, in_channels=1600))

7
configs/classification/baseline/cub/baseline_conv4_cub_5way_5shot.py
View File

@ -1,7 +0,0 @@
_base_ = ['baseline_cub_5way_5shot_84x84_aug.py']
model = dict(
type='BaselineClassifier',
backbone=dict(type='Conv4'),
head=dict(type='LinearHead', num_classes=100, in_channels=1600),
meta_test_head=dict(type='LinearHead', num_classes=5, in_channels=1600))

10
configs/classification/baseline/cub/baseline_cub_5way_1shot_84x84_aug.py → configs/classification/baseline/cub/baseline_resnet12_1xb64_cub_5way-1shot.py
View File

@ -1,5 +1,5 @@
_base_ = [
'../../_base_/meta_test/cub_meta_test_5way_1shot_84x84.py',
'../../_base_/meta_test/cub_meta-test_5way-1shot.py',
'../../_base_/runtime/epoch_based_runtime.py',
'../../_base_/schedules/sgd_200epoch.py'
]
@ -25,7 +25,7 @@ meta_finetune_cfg = dict(
data = dict(
samples_per_gpu=64,
workers_per_gpu=6,
workers_per_gpu=4,
train=dict(
type='CUBDataset',
data_prefix='data/CUB_200_2011',
@ -39,3 +39,9 @@ data = dict(
meta_test_cfg=dict(
support=dict(
batch_size=4, drop_last=True, train=meta_finetune_cfg))))
model = dict(
type='BaselineClassifier',
backbone=dict(type='ResNet12'),
head=dict(type='LinearHead', num_classes=100, in_channels=640),
meta_test_head=dict(type='LinearHead', num_classes=5, in_channels=640))

10
configs/classification/baseline/cub/baseline_cub_5way_5shot_84x84_aug.py → configs/classification/baseline/cub/baseline_resnet12_1xb64_cub_5way-5shot.py
View File

@ -1,5 +1,5 @@
_base_ = [
'../../_base_/meta_test/cub_meta_test_5way_5shot_84x84.py',
'../../_base_/meta_test/cub_meta-test_5way-5shot.py',
'../../_base_/runtime/epoch_based_runtime.py',
'../../_base_/schedules/sgd_200epoch.py'
]
@ -25,7 +25,7 @@ meta_finetune_cfg = dict(
data = dict(
samples_per_gpu=64,
workers_per_gpu=6,
workers_per_gpu=4,
train=dict(
type='CUBDataset',
data_prefix='data/CUB_200_2011',
@ -39,3 +39,9 @@ data = dict(
meta_test_cfg=dict(
support=dict(
batch_size=4, drop_last=True, train=meta_finetune_cfg))))
model = dict(
type='BaselineClassifier',
backbone=dict(type='ResNet12'),
head=dict(type='LinearHead', num_classes=100, in_channels=640),
meta_test_head=dict(type='LinearHead', num_classes=5, in_channels=640))

6
configs/classification/baseline/cub/baseline_resnet12_cub_5way_1shot.py
View File

@ -1,6 +0,0 @@
_base_ = ['baseline_cub_5way_1shot_84x84_aug.py']
model = dict(
type='BaselineClassifier',
backbone=dict(type='ResNet12'),
head=dict(type='LinearHead', num_classes=100, in_channels=640),
meta_test_head=dict(type='LinearHead', num_classes=5, in_channels=640))

6
configs/classification/baseline/cub/baseline_resnet12_cub_5way_5shot.py
View File

@ -1,6 +0,0 @@
_base_ = ['baseline_cub_5way_5shot_84x84_aug.py']
model = dict(
type='BaselineClassifier',
backbone=dict(type='ResNet12'),
head=dict(type='LinearHead', num_classes=100, in_channels=640),
meta_test_head=dict(type='LinearHead', num_classes=5, in_channels=640))

64
configs/classification/baseline/cub/metafile.yml 100644
View File

@ -0,0 +1,64 @@
Collections:
- Name: Baseline
Metadata:
Training Data: CUB
Training Techniques:
- SGD with Momentum
- Weight Decay
Training Resources: 1x V100 GPUs
Paper: https://arxiv.org/abs/1904.04232
README: configs/classification/baseline/README.md
Models:
- Name: baseline_conv4_1xb64_cub_5way-1shot
Metadata:
Training Data: CUB
Epochs: 200
Batch Size: 64
In Collection: Baseline
Results:
- Task: Few Shot Image Classification
Dataset: CUB
Metrics:
Accuracy: none
Weights: none
Config: configs/classification/baseline/cub/baseline_conv4_1xb64_cub_5way-1shot.py
- Name: baseline_conv4_1xb64_cub_5way-5shot
Metadata:
Training Data: CUB
Epochs: 200
Batch Size: 64
In Collection: Baseline
Results:
- Task: Few Shot Image Classification
Dataset: CUB
Metrics:
Accuracy: none
Weights: none
Config: configs/classification/baseline/cub/baseline_conv4_1xb64_cub_5way-5shot.py
- Name: baseline_resnet12_1xb64_cub_5way-1shot
Metadata:
Training Data: CUB
Epochs: 200
Batch Size: 64
In Collection: Baseline
Results:
- Task: Few Shot Image Classification
Dataset: CUB
Metrics:
Accuracy: none
Weights: none
Config: configs/classification/baseline/cub/baseline_resnet12_1xb64_cub_5way-1shot.py
- Name: baseline_resnet12_1xb64_cub_5way-5shot
Metadata:
Training Data: CUB
Epochs: 200
Batch Size: 64
In Collection: Baseline
Results:
- Task: Few Shot Image Classification
Dataset: CUB
Metrics:
Accuracy: none
Weights: none
Config: configs/classification/baseline/cub/baseline_resnet12_1xb64_cub_5way-5shot.py

10
configs/classification/baseline/mini_imagenet/baseline_mini_imagenet_5way_1shot_84x84_aug.py → configs/classification/baseline/mini_imagenet/baseline_conv4_1xb64_mini-imagenet_5way-1shot.py
View File

@ -1,5 +1,5 @@
_base_ = [
'../../_base_/meta_test/mini_imagenet_meta_test_5way_1shot_84x84.py',
'../../_base_/meta_test/mini-imagenet_meta-test_5way-1shot.py',
'../../_base_/runtime/epoch_based_runtime.py',
'../../_base_/schedules/sgd_200epoch.py'
]
@ -25,7 +25,7 @@ meta_finetune_cfg = dict(
data = dict(
samples_per_gpu=64,
workers_per_gpu=6,
workers_per_gpu=4,
train=dict(
type='MiniImageNetDataset',
data_prefix='data/mini_imagenet',
@ -39,3 +39,9 @@ data = dict(
meta_test_cfg=dict(
support=dict(
batch_size=4, drop_last=True, train=meta_finetune_cfg))))
model = dict(
type='BaselineClassifier',
backbone=dict(type='Conv4'),
head=dict(type='LinearHead', num_classes=64, in_channels=1600),
meta_test_head=dict(type='LinearHead', num_classes=5, in_channels=1600))

10
configs/classification/baseline/mini_imagenet/baseline_mini_imagenet_5way_5shot_84x84_aug.py → configs/classification/baseline/mini_imagenet/baseline_conv4_1xb64_mini-imagenet_5way-5shot.py
View File

@ -1,5 +1,5 @@
_base_ = [
'../../_base_/meta_test/mini_imagenet_meta_test_5way_5shot_84x84.py',
'../../_base_/meta_test/mini-imagenet_meta-test_5way-5shot.py',
'../../_base_/runtime/epoch_based_runtime.py',
'../../_base_/schedules/sgd_200epoch.py'
]
@ -25,7 +25,7 @@ meta_finetune_cfg = dict(
data = dict(
samples_per_gpu=64,
workers_per_gpu=6,
workers_per_gpu=4,
train=dict(
type='MiniImageNetDataset',
data_prefix='data/mini_imagenet',
@ -39,3 +39,9 @@ data = dict(
meta_test_cfg=dict(
support=dict(
batch_size=4, drop_last=True, train=meta_finetune_cfg))))
model = dict(
type='BaselineClassifier',
backbone=dict(type='Conv4'),
head=dict(type='LinearHead', num_classes=64, in_channels=1600),
meta_test_head=dict(type='LinearHead', num_classes=5, in_channels=1600))

6
configs/classification/baseline/mini_imagenet/baseline_conv4_mini_imagenet_5way_1shot.py
View File

@ -1,6 +0,0 @@
_base_ = ['baseline_mini_imagenet_5way_1shot_84x84_aug.py']
model = dict(
type='BaselineClassifier',
backbone=dict(type='Conv4'),
head=dict(type='LinearHead', num_classes=64, in_channels=1600),
meta_test_head=dict(type='LinearHead', num_classes=5, in_channels=1600))

6
configs/classification/baseline/mini_imagenet/baseline_conv4_mini_imagenet_5way_5shot.py
View File

@ -1,6 +0,0 @@
_base_ = ['baseline_mini_imagenet_5way_5shot_84x84_aug.py']
model = dict(
type='BaselineClassifier',
backbone=dict(type='Conv4'),
head=dict(type='LinearHead', num_classes=64, in_channels=1600),
meta_test_head=dict(type='LinearHead', num_classes=5, in_channels=1600))

10
configs/classification/baseline_plus/mini_imagenet/baseline_plus_mini_imagenet_5way_1shot_84x84_aug.py → configs/classification/baseline/mini_imagenet/baseline_resnet12_1xb64_mini-imagenet_5way-1shot.py
View File

@ -1,5 +1,5 @@
_base_ = [
'../../_base_/meta_test/mini_imagenet_meta_test_5way_1shot_84x84.py',
'../../_base_/meta_test/mini-imagenet_meta-test_5way-1shot.py',
'../../_base_/runtime/epoch_based_runtime.py',
'../../_base_/schedules/sgd_200epoch.py'
]
@ -25,7 +25,7 @@ meta_finetune_cfg = dict(
data = dict(
samples_per_gpu=64,
workers_per_gpu=6,
workers_per_gpu=4,
train=dict(
type='MiniImageNetDataset',
data_prefix='data/mini_imagenet',
@ -39,3 +39,9 @@ data = dict(
meta_test_cfg=dict(
support=dict(
batch_size=4, drop_last=True, train=meta_finetune_cfg))))
model = dict(
type='BaselineClassifier',
backbone=dict(type='ResNet12'),
head=dict(type='LinearHead', num_classes=64, in_channels=640),
meta_test_head=dict(type='LinearHead', num_classes=5, in_channels=640))

10
configs/classification/baseline_plus/mini_imagenet/baseline_plus_mini_imagenet_5way_5shot_84x84_aug.py → configs/classification/baseline/mini_imagenet/baseline_resnet12_1xb64_mini-imagenet_5way-5shot.py
View File

@ -1,5 +1,5 @@
_base_ = [
'../../_base_/meta_test/mini_imagenet_meta_test_5way_5shot_84x84.py',
'../../_base_/meta_test/mini-imagenet_meta-test_5way-5shot.py',
'../../_base_/runtime/epoch_based_runtime.py',
'../../_base_/schedules/sgd_200epoch.py'
]
@ -25,7 +25,7 @@ meta_finetune_cfg = dict(
data = dict(
samples_per_gpu=64,
workers_per_gpu=6,
workers_per_gpu=4,
train=dict(
type='MiniImageNetDataset',
data_prefix='data/mini_imagenet',
@ -39,3 +39,9 @@ data = dict(
meta_test_cfg=dict(
support=dict(
batch_size=4, drop_last=True, train=meta_finetune_cfg))))
model = dict(
type='BaselineClassifier',
backbone=dict(type='ResNet12'),
head=dict(type='LinearHead', num_classes=64, in_channels=640),
meta_test_head=dict(type='LinearHead', num_classes=5, in_channels=640))

6
configs/classification/baseline/mini_imagenet/baseline_resnet12_mini_imagenet_5way_1shot.py
View File

@ -1,6 +0,0 @@
_base_ = ['baseline_mini_imagenet_5way_1shot_84x84_aug.py']
model = dict(
type='BaselineClassifier',
backbone=dict(type='ResNet12'),
head=dict(type='LinearHead', num_classes=64, in_channels=640),
meta_test_head=dict(type='LinearHead', num_classes=5, in_channels=640))

6
configs/classification/baseline/mini_imagenet/baseline_resnet12_mini_imagenet_5way_5shot.py
View File

@ -1,6 +0,0 @@
_base_ = ['baseline_mini_imagenet_5way_5shot_84x84_aug.py']
model = dict(
type='BaselineClassifier',
backbone=dict(type='ResNet12'),
head=dict(type='LinearHead', num_classes=64, in_channels=640),
meta_test_head=dict(type='LinearHead', num_classes=5, in_channels=640))

64
configs/classification/baseline/mini_imagenet/metafile.yml 100644
View File

@ -0,0 +1,64 @@
Collections:
- Name: Baseline
Metadata:
Training Data: Mini-ImageNet
Training Techniques:
- SGD with Momentum
- Weight Decay
Training Resources: 1x V100 GPUs
Paper: https://arxiv.org/abs/1904.04232
README: configs/classification/baseline/README.md
Models:
- Name: baseline_conv4_1xb64_mini-imagenet_5way-1shot
Metadata:
Training Data: Mini-ImageNet
Epochs: 200
Batch Size: 64
In Collection: Baseline
Results:
- Task: Few Shot Image Classification
Dataset: Mini-ImageNet
Metrics:
Accuracy: none
Weights: none
Config: configs/classification/baseline/mini_imagenet/baseline_conv4_1xb64_mini-imagenet_5way-1shot.py
- Name: baseline_conv4_1xb64_mini-imagenet_5way-5shot
Metadata:
Training Data: Mini-ImageNet
Epochs: 200
Batch Size: 64
In Collection: Baseline
Results:
- Task: Few Shot Image Classification
Dataset: Mini-ImageNet
Metrics:
Accuracy: none
Weights: none
Config: configs/classification/baseline/mini_imagenet/baseline_conv4_1xb64_mini-imagenet_5way-5shot.py
- Name: baseline_resnet12_1xb64_mini-imagenet_5way-1shot
Metadata:
Training Data: Mini-ImageNet
Epochs: 200
Batch Size: 64
In Collection: Baseline
Results:
- Task: Few Shot Image Classification
Dataset: Mini-ImageNet
Metrics:
Accuracy: none
Weights: none
Config: configs/classification/baseline/mini_imagenet/baseline_resnet12_1xb64_mini-imagenet_5way-1shot.py
- Name: baseline_resnet12_1xb64_mini-imagenet_5way-5shot
Metadata:
Training Data: Mini-ImageNet
Epochs: 200
Batch Size: 64
In Collection: Baseline
Results:
- Task: Few Shot Image Classification
Dataset: Mini-ImageNet
Metrics:
Accuracy: none
Weights: none
Config: configs/classification/baseline/mini_imagenet/baseline_resnet12_1xb64_mini-imagenet_5way-5shot.py

11
configs/classification/baseline/tiered_imagenet/baseline_tiered_imagenet_5way_1shot_84x84_aug.py → configs/classification/baseline/tiered_imagenet/baseline_conv4_1xb64_tiered-imagenet_5way-1shot.py
View File

@ -1,5 +1,5 @@
_base_ = [
'../../_base_/meta_test/tiered_imagenet_meta_test_5way_1shot_84x84.py',
'../../_base_/meta_test/tiered-imagenet_meta-test_5way-1shot.py',
'../../_base_/runtime/epoch_based_runtime.py',
'../../_base_/schedules/sgd_100epoch.py'
]
@ -25,7 +25,7 @@ meta_finetune_cfg = dict(
data = dict(
samples_per_gpu=64,
workers_per_gpu=8,
workers_per_gpu=4,
train=dict(
type='TieredImageNetDataset',
data_prefix='data/tiered_imagenet',
@ -39,4 +39,9 @@ data = dict(
meta_test_cfg=dict(
support=dict(
batch_size=4, drop_last=True, train=meta_finetune_cfg))))
pin_memory = True
model = dict(
type='BaselineClassifier',
backbone=dict(type='Conv4'),
head=dict(type='LinearHead', num_classes=351, in_channels=1600),
meta_test_head=dict(type='LinearHead', num_classes=5, in_channels=1600))

11
configs/classification/baseline/tiered_imagenet/baseline_tiered_imagenet_5way_5shot_84x84_aug.py → configs/classification/baseline/tiered_imagenet/baseline_conv4_1xb64_tiered-imagenet_5way-5shot.py
View File

@ -1,5 +1,5 @@
_base_ = [
'../../_base_/meta_test/tiered_imagenet_meta_test_5way_5shot_84x84.py',
'../../_base_/meta_test/tiered-imagenet_meta-test_5way-5shot.py',
'../../_base_/runtime/epoch_based_runtime.py',
'../../_base_/schedules/sgd_100epoch.py'
]
@ -25,7 +25,7 @@ meta_finetune_cfg = dict(
data = dict(
samples_per_gpu=64,
workers_per_gpu=8,
workers_per_gpu=4,
train=dict(
type='TieredImageNetDataset',
data_prefix='data/tiered_imagenet',
@ -39,4 +39,9 @@ data = dict(
meta_test_cfg=dict(
support=dict(
batch_size=4, drop_last=True, train=meta_finetune_cfg))))
pin_memory = True
model = dict(
type='BaselineClassifier',
backbone=dict(type='Conv4'),
head=dict(type='LinearHead', num_classes=351, in_channels=1600),
meta_test_head=dict(type='LinearHead', num_classes=5, in_channels=1600))

7
configs/classification/baseline/tiered_imagenet/baseline_conv4_tiered_imagenet_5way_1shot.py
View File

@ -1,7 +0,0 @@
_base_ = ['baseline_tiered_imagenet_5way_1shot_84x84_aug.py']
model = dict(
type='BaselineClassifier',
backbone=dict(type='Conv4'),
head=dict(type='LinearHead', num_classes=351, in_channels=1600),
meta_test_head=dict(type='LinearHead', num_classes=5, in_channels=1600))

7
configs/classification/baseline/tiered_imagenet/baseline_conv4_tiered_imagenet_5way_5shot.py
View File

@ -1,7 +0,0 @@
_base_ = ['baseline_tiered_imagenet_5way_5shot_84x84_aug.py']
model = dict(
type='BaselineClassifier',
backbone=dict(type='Conv4'),
head=dict(type='LinearHead', num_classes=351, in_channels=1600),
meta_test_head=dict(type='LinearHead', num_classes=5, in_channels=1600))

47
configs/classification/baseline/tiered_imagenet/baseline_resnet12_1xb64_tiered-imagenet_5way-1shot.py 100644
View File

@ -0,0 +1,47 @@
_base_ = [
'../../_base_/meta_test/tiered-imagenet_meta-test_5way-1shot.py',
'../../_base_/runtime/epoch_based_runtime.py',
'../../_base_/schedules/sgd_100epoch.py'
]
img_size = 84
img_norm_cfg = dict(
mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
train_pipeline = [
dict(type='LoadImageFromBytes'),
dict(type='RandomResizedCrop', size=img_size),
dict(type='RandomFlip', flip_prob=0.5, direction='horizontal'),
dict(type='ColorJitter', brightness=0.4, contrast=0.4, saturation=0.4),
dict(type='Normalize', **img_norm_cfg),
dict(type='ImageToTensor', keys=['img']),
dict(type='ToTensor', keys=['gt_label']),
dict(type='Collect', keys=['img', 'gt_label'])
]
meta_finetune_cfg = dict(
num_steps=150,
optimizer=dict(
type='SGD', lr=0.01, momentum=0.9, dampening=0.9, weight_decay=0.001))
data = dict(
samples_per_gpu=64,
workers_per_gpu=4,
train=dict(
type='TieredImageNetDataset',
data_prefix='data/tiered_imagenet',
subset='train',
pipeline=train_pipeline),
val=dict(
meta_test_cfg=dict(
support=dict(
batch_size=4, drop_last=True, train=meta_finetune_cfg))),
test=dict(
meta_test_cfg=dict(
support=dict(
batch_size=4, drop_last=True, train=meta_finetune_cfg))))
model = dict(
type='BaselineClassifier',
backbone=dict(type='ResNet12'),
head=dict(type='LinearHead', num_classes=351, in_channels=640),
meta_test_head=dict(type='LinearHead', num_classes=5, in_channels=640))

47
configs/classification/baseline/tiered_imagenet/baseline_resnet12_1xb64_tiered-imagenet_5way-5shot.py 100644
View File

@ -0,0 +1,47 @@
_base_ = [
'../../_base_/meta_test/tiered-imagenet_meta-test_5way-5shot.py',
'../../_base_/runtime/epoch_based_runtime.py',
'../../_base_/schedules/sgd_100epoch.py'
]
img_size = 84
img_norm_cfg = dict(
mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
train_pipeline = [
dict(type='LoadImageFromBytes'),
dict(type='RandomResizedCrop', size=img_size),
dict(type='RandomFlip', flip_prob=0.5, direction='horizontal'),
dict(type='ColorJitter', brightness=0.4, contrast=0.4, saturation=0.4),
dict(type='Normalize', **img_norm_cfg),
dict(type='ImageToTensor', keys=['img']),
dict(type='ToTensor', keys=['gt_label']),
dict(type='Collect', keys=['img', 'gt_label'])
]
meta_finetune_cfg = dict(
num_steps=600,
optimizer=dict(
type='SGD', lr=0.01, momentum=0.9, dampening=0.9, weight_decay=0.001))
data = dict(
samples_per_gpu=64,
workers_per_gpu=4,
train=dict(
type='TieredImageNetDataset',
data_prefix='data/tiered_imagenet',
subset='train',
pipeline=train_pipeline),
val=dict(
meta_test_cfg=dict(
support=dict(
batch_size=4, drop_last=True, train=meta_finetune_cfg))),
test=dict(
meta_test_cfg=dict(
support=dict(
batch_size=4, drop_last=True, train=meta_finetune_cfg))))
model = dict(
type='BaselineClassifier',
backbone=dict(type='ResNet12'),
head=dict(type='LinearHead', num_classes=351, in_channels=640),
meta_test_head=dict(type='LinearHead', num_classes=5, in_channels=640))

7
configs/classification/baseline/tiered_imagenet/baseline_resnet12_tiered_imagenet_5way_1shot.py
View File

@ -1,7 +0,0 @@
_base_ = ['baseline_tiered_imagenet_5way_1shot_84x84_aug.py']
model = dict(
type='BaselineClassifier',
backbone=dict(type='ResNet12'),
head=dict(type='LinearHead', num_classes=351, in_channels=640),
meta_test_head=dict(type='LinearHead', num_classes=5, in_channels=640))

7
configs/classification/baseline/tiered_imagenet/baseline_resnet12_tiered_imagenet_5way_5shot.py
View File

@ -1,7 +0,0 @@
_base_ = ['baseline_tiered_imagenet_5way_5shot_84x84_aug.py']
model = dict(
type='BaselineClassifier',
backbone=dict(type='ResNet12'),
head=dict(type='LinearHead', num_classes=351, in_channels=640),
meta_test_head=dict(type='LinearHead', num_classes=5, in_channels=640))

64
configs/classification/baseline/tiered_imagenet/metafile.yml 100644
View File

@ -0,0 +1,64 @@
Collections:
- Name: Baseline
Metadata:
Training Data: Tiered-ImageNet
Training Techniques:
- SGD with Momentum
- Weight Decay
Training Resources: 1x V100 GPUs
Paper: https://arxiv.org/abs/1904.04232
README: configs/classification/baseline/README.md
Models:
- Name: baseline_conv4_1xb64_tiered-imagenet_5way-1shot
Metadata:
Training Data: Tiered-ImageNet
Epochs: 100
Batch Size: 64
In Collection: Baseline
Results:
- Task: Few Shot Image Classification
Dataset: Tiered-ImageNet
Metrics:
Accuracy: none
Weights: none
Config: configs/classification/baseline/tiered_imagenet/baseline_conv4_1xb64_tiered-imagenet_5way-1shot.py
- Name: baseline_conv4_1xb64_tiered-imagenet_5way-5shot
Metadata:
Training Data: Tiered-ImageNet
Epochs: 100
Batch Size: 64
In Collection: Baseline
Results:
- Task: Few Shot Image Classification
Dataset: Tiered-ImageNet
Metrics:
Accuracy: none
Weights: none
Config: configs/classification/baseline/tiered_imagenet/baseline_conv4_1xb64_tiered-imagenet_5way-5shot.py
- Name: baseline_resnet12_1xb64_tiered-imagenet_5way-1shot
Metadata:
Training Data: Tiered-ImageNet
Epochs: 100
Batch Size: 64
In Collection: Baseline
Results:
- Task: Few Shot Image Classification
Dataset: Tiered-ImageNet
Metrics:
Accuracy: none
Weights: none
Config: configs/classification/baseline/tiered_imagenet/baseline_resnet12_1xb64_tiered-imagenet_5way-1shot.py
- Name: baseline_resnet12_1xb64_tiered-imagenet_5way-5shot
Metadata:
Training Data: Tiered-ImageNet
Epochs: 100
Batch Size: 64
In Collection: Baseline
Results:
- Task: Few Shot Image Classification
Dataset: Tiered-ImageNet
Metrics:
Accuracy: none
Weights: none
Config: configs/classification/baseline/tiered_imagenet/baseline_resnet12_1xb64_tiered-imagenet_5way-5shot.py

67
configs/classification/baseline_plus/README.md 100644
View File

@ -0,0 +1,67 @@
<!-- [ALGORITHM] -->
#<summary><a href="https://arxiv.org/abs/1904.04232"> Baseline++ (ICLR'2019)</a></summary>
```bibtex
@inproceedings{chen2019closerfewshot,
title={A Closer Look at Few-shot Classification},
author={Chen, Wei-Yu and Liu, Yen-Cheng and Kira, Zsolt and Wang, Yu-Chiang and Huang, Jia-Bin},
booktitle={International Conference on Learning Representations},
year={2019}
}
```
## How to Reproduce Baseline++
It consists of two steps:
- **Step1: Base training**
- use all the images of base classes to train a base model.
- use validation set to select the best model.
- **Step2: Meta Testing**:
- use best model from step1.
### An example of CUB dataset with Conv4
```bash
# base training
python ./tools/classification/train.py \
configs/classification/baseline_plus/cub/baseline-plus_conv4_1xb64_cub_5way-1shot.py
# meta testing
python ./tools/classification/test.py \
configs/classification/baseline_plus/cub/baseline-plus_conv4_1xb64_cub_5way-1shot.py \
work_dir/baseline-plus_conv4_1xb64_cub_5way-1shot/best_accuracy_mean.pth
```
**Note**:
- All the result are trained with single gpu.
- The base training of 1 shot and 5 shot use same training setting,
but different validation setting.
## Results on CUB dataset of 1000 episodes
| Arch | Input Size | Batch Size | way | shot | mean Acc | std | ckpt | log |
| :-------------- | :-----------: | :------: | :------: | :------: | :------: | :------: |:------: |:------: |
| [conv4](/configs/classification/baseline_plus/cub/baseline-plus_conv4_1xb64_cub_5way-1shot.py) | 84x84 | 64 | 5 | 1 | - | - | [ckpt]() | [log]() |
| [conv4](/configs/classification/baseline_plus/cub/baseline-plus_conv4_1xb64_cub_5way-5shot.py) | 84x84 | 64 | 5 | 5 | - | - | [ckpt]() | [log]() |
| [resnet12](/configs/classification/baseline_plus/cub/baseline-plus_resnet12_1xb64_cub_5way-1shot.py) | 84x84 | 64 | 5 | 1 | - | - | [ckpt]() | [log]() |
| [resnet12](/configs/classification/baseline_plus/cub/baseline-plus_resnet12_1xb64_cub_5way-5shot.py) | 84x84 | 64 | 5 | 5 | - | - | [ckpt]() | [log]() |
## Results on Mini-ImageNet dataset of 1000 episodes
| Arch | Input Size | Batch Size | way | shot | mean Acc | std | ckpt | log |
| :-------------- | :-----------: | :------: | :------: | :------: | :------: | :------: |:------: |:------: |
| [conv4](/configs/classification/baseline_plus/mini_imagenet/baseline-plus_conv4_1xb64_mini-imagenet_5way-1shot.py) | 84x84 | 64 | 5 | 1 | - | - | [ckpt]() | [log]() |
| [conv4](/configs/classification/baseline_plus/mini_imagenet/baseline-plus_conv4_1xb64_mini-imagenet_5way-5shot.py) | 84x84 | 64 | 5 | 5 | - | - | [ckpt]() | [log]() |
| [resnet12](/configs/classification/baseline_plus/mini_imagenet/baseline-plus_resnet12_1xb64_mini-imagenet_5way-1shot.py) | 84x84 | 64 | 5 | 1 | - | - | [ckpt]() | [log]() |
| [resnet12](/configs/classification/baseline_plus/mini_imagenet/baseline-plus_resnet12_1xb64_mini-imagenet_5way-5shot.py) | 84x84 | 64 | 5 | 5 | - | - | [ckpt]() | [log]() |
## Results on Tiered-ImageNet of 1000 episodes
| Arch | Input Size | Batch Size | way | shot | mean Acc | std | ckpt | log |
| :-------------- | :-----------: | :------: | :------: | :------: | :------: | :------: |:------: |:------: |
| [conv4](/configs/classification/baseline_plus/tiered_imagenet/baseline-plus_conv4_1xb64_tiered-imagenet_5way-1shot.py) | 84x84 | 64 | 5 | 1 | - | - | [ckpt]() | [log]() |
| [conv4](/configs/classification/baseline_plus/tiered_imagenet/baseline-plus_conv4_1xb64_tiered-imagenet_5way-5shot.py) | 84x84 | 64 | 5 | 5 | - | - | [ckpt]() | [log]() |
| [resnet12](/configs/classification/baseline_plus/tiered_imagenet/baseline-plus_resnet12_1xb64_tiered-imagenet_5way-1shot.py) | 84x84 | 64 | 5 | 1 | - | - | [ckpt]() | [log]() |
| [resnet12](/configs/classification/baseline_plus/tiered_imagenet/baseline-plus_resnet12_1xb64_tiered-imagenet_5way-5shot.py) | 84x84 | 64 | 5 | 5 | - | - | [ckpt]() | [log]() |

48
configs/classification/baseline_plus/cub/baseline-plus_conv4_1xb64_cub_5way-1shot.py 100644
View File

@ -0,0 +1,48 @@
_base_ = [
'../../_base_/meta_test/cub_meta-test_5way-1shot.py',
'../../_base_/runtime/epoch_based_runtime.py',
'../../_base_/schedules/sgd_200epoch.py'
]
img_size = 84
img_norm_cfg = dict(
mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='RandomResizedCrop', size=img_size),
dict(type='RandomFlip', flip_prob=0.5, direction='horizontal'),
dict(type='ColorJitter', brightness=0.4, contrast=0.4, saturation=0.4),
dict(type='Normalize', **img_norm_cfg),
dict(type='ImageToTensor', keys=['img']),
dict(type='ToTensor', keys=['gt_label']),
dict(type='Collect', keys=['img', 'gt_label'])
]
meta_finetune_cfg = dict(
num_steps=150,
optimizer=dict(
type='SGD', lr=0.01, momentum=0.9, dampening=0.9, weight_decay=0.001))
data = dict(
samples_per_gpu=64,
workers_per_gpu=4,
train=dict(
type='CUBDataset',
data_prefix='data/CUB_200_2011',
subset='train',
pipeline=train_pipeline),
val=dict(
meta_test_cfg=dict(
support=dict(
batch_size=4, drop_last=True, train=meta_finetune_cfg))),
test=dict(
meta_test_cfg=dict(
support=dict(
batch_size=4, drop_last=True, train=meta_finetune_cfg))))
model = dict(
type='BaselinePlusClassifier',
backbone=dict(type='Conv4'),
head=dict(type='CosineDistanceHead', num_classes=100, in_channels=1600),
meta_test_head=dict(
type='CosineDistanceHead', num_classes=5, in_channels=1600))

48
configs/classification/baseline_plus/cub/baseline-plus_conv4_1xb64_cub_5way-5shot.py 100644
View File

@ -0,0 +1,48 @@
_base_ = [
'../../_base_/meta_test/cub_meta-test_5way-5shot.py',
'../../_base_/runtime/epoch_based_runtime.py',
'../../_base_/schedules/sgd_200epoch.py'
]
img_size = 84
img_norm_cfg = dict(
mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='RandomResizedCrop', size=img_size),
dict(type='RandomFlip', flip_prob=0.5, direction='horizontal'),
dict(type='ColorJitter', brightness=0.4, contrast=0.4, saturation=0.4),
dict(type='Normalize', **img_norm_cfg),
dict(type='ImageToTensor', keys=['img']),
dict(type='ToTensor', keys=['gt_label']),
dict(type='Collect', keys=['img', 'gt_label'])
]
meta_finetune_cfg = dict(
num_steps=600,
optimizer=dict(
type='SGD', lr=0.01, momentum=0.9, dampening=0.9, weight_decay=0.001))
data = dict(
samples_per_gpu=64,
workers_per_gpu=4,
train=dict(
type='CUBDataset',
data_prefix='data/CUB_200_2011',
subset='train',
pipeline=train_pipeline),
val=dict(
meta_test_cfg=dict(
support=dict(
batch_size=4, drop_last=True, train=meta_finetune_cfg))),
test=dict(
meta_test_cfg=dict(
support=dict(
batch_size=4, drop_last=True, train=meta_finetune_cfg))))
model = dict(
type='BaselinePlusClassifier',
backbone=dict(type='Conv4'),
head=dict(type='CosineDistanceHead', num_classes=100, in_channels=1600),
meta_test_head=dict(
type='CosineDistanceHead', num_classes=5, in_channels=1600))

55
configs/classification/baseline_plus/cub/baseline-plus_resnet12_1xb64_cub_5way-1shot.py 100644
View File

@ -0,0 +1,55 @@
_base_ = [
'../../_base_/meta_test/cub_meta-test_5way-1shot.py',
'../../_base_/runtime/epoch_based_runtime.py',
'../../_base_/schedules/sgd_200epoch.py'
]
img_size = 84
img_norm_cfg = dict(
mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='RandomResizedCrop', size=img_size),
dict(type='RandomFlip', flip_prob=0.5, direction='horizontal'),
dict(type='ColorJitter', brightness=0.4, contrast=0.4, saturation=0.4),
dict(type='Normalize', **img_norm_cfg),
dict(type='ImageToTensor', keys=['img']),
dict(type='ToTensor', keys=['gt_label']),
dict(type='Collect', keys=['img', 'gt_label'])
]
meta_finetune_cfg = dict(
num_steps=150,
optimizer=dict(
type='SGD', lr=0.01, momentum=0.9, dampening=0.9, weight_decay=0.001))
data = dict(
samples_per_gpu=64,
workers_per_gpu=4,
train=dict(
type='CUBDataset',
data_prefix='data/CUB_200_2011',
subset='train',
pipeline=train_pipeline),
val=dict(
meta_test_cfg=dict(
support=dict(
batch_size=4, drop_last=True, train=meta_finetune_cfg))),
test=dict(
meta_test_cfg=dict(
support=dict(
batch_size=4, drop_last=True, train=meta_finetune_cfg))))
model = dict(
type='BaselinePlusClassifier',
backbone=dict(type='ResNet12'),
head=dict(
type='CosineDistanceHead',
num_classes=100,
in_channels=640,
temperature=10.0),
meta_test_head=dict(
type='CosineDistanceHead',
num_classes=5,
in_channels=640,
temperature=5.0))

55
configs/classification/baseline_plus/cub/baseline-plus_resnet12_1xb64_cub_5way-5shot.py 100644
View File

@ -0,0 +1,55 @@
_base_ = [
'../../_base_/meta_test/cub_meta-test_5way-5shot.py',
'../../_base_/runtime/epoch_based_runtime.py',
'../../_base_/schedules/sgd_200epoch.py'
]
img_size = 84
img_norm_cfg = dict(
mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='RandomResizedCrop', size=img_size),
dict(type='RandomFlip', flip_prob=0.5, direction='horizontal'),
dict(type='ColorJitter', brightness=0.4, contrast=0.4, saturation=0.4),
dict(type='Normalize', **img_norm_cfg),
dict(type='ImageToTensor', keys=['img']),
dict(type='ToTensor', keys=['gt_label']),
dict(type='Collect', keys=['img', 'gt_label'])
]
meta_finetune_cfg = dict(
num_steps=600,
optimizer=dict(
type='SGD', lr=0.01, momentum=0.9, dampening=0.9, weight_decay=0.001))
data = dict(
samples_per_gpu=64,
workers_per_gpu=4,
train=dict(
type='CUBDataset',
data_prefix='data/CUB_200_2011',
subset='train',
pipeline=train_pipeline),
val=dict(
meta_test_cfg=dict(
support=dict(
batch_size=4, drop_last=True, train=meta_finetune_cfg))),
test=dict(
meta_test_cfg=dict(
support=dict(
batch_size=4, drop_last=True, train=meta_finetune_cfg))))
model = dict(
type='BaselinePlusClassifier',
backbone=dict(type='ResNet12'),
head=dict(
type='CosineDistanceHead',
num_classes=100,
in_channels=640,
temperature=10.0),
meta_test_head=dict(
type='CosineDistanceHead',
num_classes=5,
in_channels=640,
temperature=5.0))

8
configs/classification/baseline_plus/cub/baseline_plus_conv4_cub_5way_1shot.py
View File

@ -1,8 +0,0 @@
_base_ = ['baseline_pp_cub_5way_1shot_84x84_aug.py']
model = dict(
type='BaselinePlusClassifier',
backbone=dict(type='Conv4'),
head=dict(type='CosineDistanceHead', num_classes=100, in_channels=1600),
meta_test_head=dict(
type='CosineDistanceHead', num_classes=5, in_channels=1600))

8
configs/classification/baseline_plus/cub/baseline_plus_conv4_cub_5way_5shot.py
View File

@ -1,8 +0,0 @@
_base_ = ['baseline_pp_cub_5way_5shot_84x84_aug.py']
model = dict(
type='BaselinePlusClassifier',
backbone=dict(type='Conv4'),
head=dict(type='CosineDistanceHead', num_classes=100, in_channels=1600),
meta_test_head=dict(
type='CosineDistanceHead', num_classes=5, in_channels=1600))

15
configs/classification/baseline_plus/cub/baseline_plus_resnet12_cub_5way_1shot.py
View File

@ -1,15 +0,0 @@
_base_ = ['baseline_pp_cub_5way_1shot_84x84_aug.py']
model = dict(
type='BaselinePlusClassifier',
backbone=dict(type='ResNet12'),
head=dict(
type='CosineDistanceHead',
num_classes=100,
in_channels=640,
temperature=10.0),
meta_test_head=dict(
type='CosineDistanceHead',
num_classes=5,
in_channels=640,
temperature=5.0))

15
configs/classification/baseline_plus/cub/baseline_plus_resnet12_cub_5way_5shot.py
View File

@ -1,15 +0,0 @@
_base_ = ['baseline_pp_cub_5way_5shot_84x84_aug.py']
model = dict(
type='BaselinePlusClassifier',
backbone=dict(type='ResNet12'),
head=dict(
type='CosineDistanceHead',
num_classes=100,
in_channels=640,
temperature=10.0),
meta_test_head=dict(
type='CosineDistanceHead',
num_classes=5,
in_channels=640,
temperature=5.0))

64
configs/classification/baseline_plus/cub/metafile.yml 100644
View File

@ -0,0 +1,64 @@
Collections:
- Name: Baseline_Plus
Metadata:
Training Data: CUB
Training Techniques:
- SGD with Momentum
- Weight Decay
Training Resources: 1x V100 GPUs
Paper: https://arxiv.org/abs/1904.04232
README: configs/classification/baseline_plus/README.md
Models:
- Name: baseline-plus_conv4_1xb64_cub_5way-1shot
Metadata:
Training Data: CUB
Epochs: 200
Batch Size: 64
In Collection: Baseline_Plus
Results:
- Task: Few Shot Image Classification
Dataset: CUB
Metrics:
Accuracy: none
Weights: none
Config: configs/classification/baseline_plus/cub/baseline-plus_conv4_1xb64_cub_5way-1shot.py
- Name: baseline-plus_conv4_1xb64_cub_5way-5shot
Metadata:
Training Data: CUB
Epochs: 200
Batch Size: 64
In Collection: Baseline_Plus
Results:
- Task: Few Shot Image Classification
Dataset: CUB
Metrics:
Accuracy: none
Weights: none
Config: configs/classification/baseline_plus/cub/baseline-plus_conv4_1xb64_cub_5way-5shot.py
- Name: baseline-plus_resnet12_1xb64_cub_5way-1shot
Metadata:
Training Data: CUB
Epochs: 200
Batch Size: 64
In Collection: Baseline_Plus
Results:
- Task: Few Shot Image Classification
Dataset: CUB
Metrics:
Accuracy: none
Weights: none
Config: configs/classification/baseline_plus/cub/baseline-plus_resnet12_1xb64_cub_5way-1shot.py
- Name: baseline-plus_resnet12_1xb64_cub_5way-5shot
Metadata:
Training Data: CUB
Epochs: 200
Batch Size: 64
In Collection: Baseline_Plus
Results:
- Task: Few Shot Image Classification
Dataset: CUB
Metrics:
Accuracy: none
Weights: none
Config: configs/classification/baseline_plus/cub/baseline-plus_resnet12_1xb64_cub_5way-5shot.py

48
configs/classification/baseline_plus/mini_imagenet/baseline-plus_conv4_1xb64_mini-imagenet_5way-1shot.py 100644
View File

@ -0,0 +1,48 @@
_base_ = [
'../../_base_/meta_test/mini-imagenet_meta-test_5way-1shot.py',
'../../_base_/runtime/epoch_based_runtime.py',
'../../_base_/schedules/sgd_200epoch.py'
]
img_size = 84
img_norm_cfg = dict(
mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='RandomResizedCrop', size=img_size),
dict(type='RandomFlip', flip_prob=0.5, direction='horizontal'),
dict(type='ColorJitter', brightness=0.4, contrast=0.4, saturation=0.4),
dict(type='Normalize', **img_norm_cfg),
dict(type='ImageToTensor', keys=['img']),
dict(type='ToTensor', keys=['gt_label']),
dict(type='Collect', keys=['img', 'gt_label'])
]
meta_finetune_cfg = dict(
num_steps=150,
optimizer=dict(
type='SGD', lr=0.01, momentum=0.9, dampening=0.9, weight_decay=0.001))
data = dict(
samples_per_gpu=64,
workers_per_gpu=4,
train=dict(
type='MiniImageNetDataset',
data_prefix='data/mini_imagenet',
subset='train',
pipeline=train_pipeline),
val=dict(
meta_test_cfg=dict(
support=dict(
batch_size=4, drop_last=True, train=meta_finetune_cfg))),
test=dict(
meta_test_cfg=dict(
support=dict(
batch_size=4, drop_last=True, train=meta_finetune_cfg))))
model = dict(
type='BaselinePlusClassifier',
backbone=dict(type='Conv4'),
head=dict(type='CosineDistanceHead', num_classes=100, in_channels=1600),
meta_test_head=dict(
type='CosineDistanceHead', num_classes=5, in_channels=1600))

48
configs/classification/baseline_plus/mini_imagenet/baseline-plus_conv4_1xb64_mini-imagenet_5way-5shot.py 100644
View File

@ -0,0 +1,48 @@
_base_ = [
'../../_base_/meta_test/mini-imagenet_meta-test_5way-5shot.py',
'../../_base_/runtime/epoch_based_runtime.py',
'../../_base_/schedules/sgd_200epoch.py'
]
img_size = 84
img_norm_cfg = dict(
mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='RandomResizedCrop', size=img_size),
dict(type='RandomFlip', flip_prob=0.5, direction='horizontal'),
dict(type='ColorJitter', brightness=0.4, contrast=0.4, saturation=0.4),
dict(type='Normalize', **img_norm_cfg),
dict(type='ImageToTensor', keys=['img']),
dict(type='ToTensor', keys=['gt_label']),
dict(type='Collect', keys=['img', 'gt_label'])
]
meta_finetune_cfg = dict(
num_steps=600,
optimizer=dict(
type='SGD', lr=0.01, momentum=0.9, dampening=0.9, weight_decay=0.001))
data = dict(
samples_per_gpu=64,
workers_per_gpu=4,
train=dict(
type='MiniImageNetDataset',
data_prefix='data/mini_imagenet',
subset='train',
pipeline=train_pipeline),
val=dict(
meta_test_cfg=dict(
support=dict(
batch_size=4, drop_last=True, train=meta_finetune_cfg))),
test=dict(
meta_test_cfg=dict(
support=dict(
batch_size=4, drop_last=True, train=meta_finetune_cfg))))
model = dict(
type='BaselinePlusClassifier',
backbone=dict(type='Conv4'),
head=dict(type='CosineDistanceHead', num_classes=100, in_channels=1600),
meta_test_head=dict(
type='CosineDistanceHead', num_classes=5, in_channels=1600))

55
configs/classification/baseline_plus/mini_imagenet/baseline-plus_resnet12_1xb64_mini-imagenet_5way-1shot.py 100644
View File

@ -0,0 +1,55 @@
_base_ = [
'../../_base_/meta_test/mini-imagenet_meta-test_5way-1shot.py',
'../../_base_/runtime/epoch_based_runtime.py',
'../../_base_/schedules/sgd_200epoch.py'
]
img_size = 84
img_norm_cfg = dict(
mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='RandomResizedCrop', size=img_size),
dict(type='RandomFlip', flip_prob=0.5, direction='horizontal'),
dict(type='ColorJitter', brightness=0.4, contrast=0.4, saturation=0.4),
dict(type='Normalize', **img_norm_cfg),
dict(type='ImageToTensor', keys=['img']),
dict(type='ToTensor', keys=['gt_label']),
dict(type='Collect', keys=['img', 'gt_label'])
]
meta_finetune_cfg = dict(
num_steps=150,
optimizer=dict(
type='SGD', lr=0.01, momentum=0.9, dampening=0.9, weight_decay=0.001))
data = dict(
samples_per_gpu=64,
workers_per_gpu=4,
train=dict(
type='MiniImageNetDataset',
data_prefix='data/mini_imagenet',
subset='train',
pipeline=train_pipeline),
val=dict(
meta_test_cfg=dict(
support=dict(
batch_size=4, drop_last=True, train=meta_finetune_cfg))),
test=dict(
meta_test_cfg=dict(
support=dict(
batch_size=4, drop_last=True, train=meta_finetune_cfg))))
model = dict(
type='BaselinePlusClassifier',
backbone=dict(type='ResNet12'),
head=dict(
type='CosineDistanceHead',
num_classes=100,
in_channels=640,
temperature=10.0),
meta_test_head=dict(
type='CosineDistanceHead',
num_classes=5,
in_channels=640,
temperature=5.0))

55
configs/classification/baseline_plus/mini_imagenet/baseline-plus_resnet12_1xb64_mini-imagenet_5way-5shot.py 100644
View File

@ -0,0 +1,55 @@
_base_ = [
'../../_base_/meta_test/mini-imagenet_meta-test_5way-5shot.py',
'../../_base_/runtime/epoch_based_runtime.py',
'../../_base_/schedules/sgd_200epoch.py'
]
img_size = 84
img_norm_cfg = dict(
mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='RandomResizedCrop', size=img_size),
dict(type='RandomFlip', flip_prob=0.5, direction='horizontal'),
dict(type='ColorJitter', brightness=0.4, contrast=0.4, saturation=0.4),
dict(type='Normalize', **img_norm_cfg),
dict(type='ImageToTensor', keys=['img']),
dict(type='ToTensor', keys=['gt_label']),
dict(type='Collect', keys=['img', 'gt_label'])
]
meta_finetune_cfg = dict(
num_steps=600,
optimizer=dict(
type='SGD', lr=0.01, momentum=0.9, dampening=0.9, weight_decay=0.001))
data = dict(
samples_per_gpu=64,
workers_per_gpu=4,
train=dict(
type='MiniImageNetDataset',
data_prefix='data/mini_imagenet',
subset='train',
pipeline=train_pipeline),
val=dict(
meta_test_cfg=dict(
support=dict(
batch_size=4, drop_last=True, train=meta_finetune_cfg))),
test=dict(
meta_test_cfg=dict(
support=dict(
batch_size=4, drop_last=True, train=meta_finetune_cfg))))
model = dict(
type='BaselinePlusClassifier',
backbone=dict(type='ResNet12'),
head=dict(
type='CosineDistanceHead',
num_classes=100,
in_channels=640,
temperature=10.0),
meta_test_head=dict(
type='CosineDistanceHead',
num_classes=5,
in_channels=640,
temperature=5.0))

8
configs/classification/baseline_plus/mini_imagenet/baseline_plus_conv4_mini_imagenet_5way_1shot.py
View File

@ -1,8 +0,0 @@
_base_ = ['baseline_pp_mini_imagenet_5way_1shot_84x84_aug.py']
model = dict(
type='BaselinePlusClassifier',
backbone=dict(type='Conv4'),
head=dict(type='CosineDistanceHead', num_classes=100, in_channels=1600),
meta_test_head=dict(
type='CosineDistanceHead', num_classes=5, in_channels=1600))

8
configs/classification/baseline_plus/mini_imagenet/baseline_plus_conv4_mini_imagenet_5way_5shot.py
View File

@ -1,8 +0,0 @@
_base_ = ['baseline_pp_mini_imagenet_5way_5shot_84x84_aug.py']
model = dict(
type='BaselinePlusClassifier',
backbone=dict(type='Conv4'),
head=dict(type='CosineDistanceHead', num_classes=100, in_channels=1600),
meta_test_head=dict(
type='CosineDistanceHead', num_classes=5, in_channels=1600))

15
configs/classification/baseline_plus/mini_imagenet/baseline_plus_resnet12_mini_imagenet_5way_1shot.py
View File

@ -1,15 +0,0 @@
_base_ = ['baseline_pp_mini_imagenet_5way_1shot_84x84_aug.py']
model = dict(
type='BaselinePlusClassifier',
backbone=dict(type='ResNet12'),
head=dict(
type='CosineDistanceHead',
num_classes=100,
in_channels=640,
temperature=10.0),
meta_test_head=dict(
type='CosineDistanceHead',
num_classes=5,
in_channels=640,
temperature=5.0))

15
configs/classification/baseline_plus/mini_imagenet/baseline_plus_resnet12_mini_imagenet_5way_5shot.py
View File

@ -1,15 +0,0 @@
_base_ = ['baseline_pp_mini_imagenet_5way_5shot_84x84_aug.py']
model = dict(
type='BaselinePlusClassifier',
backbone=dict(type='ResNet12'),
head=dict(
type='CosineDistanceHead',
num_classes=100,
in_channels=640,
temperature=10.0),
meta_test_head=dict(
type='CosineDistanceHead',
num_classes=5,
in_channels=640,
temperature=5.0))

64
configs/classification/baseline_plus/mini_imagenet/metafile.yml 100644
View File

@ -0,0 +1,64 @@
Collections:
- Name: Baseline_Plus
Metadata:
Training Data: Mini-ImageNet
Training Techniques:
- SGD with Momentum
- Weight Decay
Training Resources: 1x V100 GPUs
Paper: https://arxiv.org/abs/1904.04232
README: configs/classification/baseline_plus/README.md
Models:
- Name: baseline-plus_conv4_1xb64_mini-imagenet_5way-1shot
Metadata:
Training Data: Mini-ImageNet
Epochs: 200
Batch Size: 64
In Collection: Baseline_Plus
Results:
- Task: Few Shot Image Classification
Dataset: Mini-ImageNet
Metrics:
Accuracy: none
Weights: none
Config: configs/classification/baseline_plus/mini_imagenet/baseline-plus_conv4_1xb64_mini-imagenet_5way-1shot.py
- Name: baseline-plus_conv4_1xb64_mini-imagenet_5way-5shot
Metadata:
Training Data: Mini-ImageNet
Epochs: 200
Batch Size: 64
In Collection: Baseline_Plus
Results:
- Task: Few Shot Image Classification
Dataset: Mini-ImageNet
Metrics:
Accuracy: none
Weights: none
Config: configs/classification/baseline_plus/mini_imagenet/baseline-plus_conv4_1xb64_mini-imagenet_5way-5shot.py
- Name: baseline-plus_resnet12_1xb64_mini-imagenet_5way-1shot
Metadata:
Training Data: Mini-ImageNet
Epochs: 200
Batch Size: 64
In Collection: Baseline_Plus
Results:
- Task: Few Shot Image Classification
Dataset: Mini-ImageNet
Metrics:
Accuracy: none
Weights: none
Config: configs/classification/baseline_plus/mini_imagenet/baseline-plus_resnet12_1xb64_mini-imagenet_5way-1shot.py
- Name: baseline-plus_resnet12_1xb64_mini-imagenet_5way-5shot
Metadata:
Training Data: Mini-ImageNet
Epochs: 200
Batch Size: 64
In Collection: Baseline_Plus
Results:
- Task: Few Shot Image Classification
Dataset: Mini-ImageNet
Metrics:
Accuracy: none
Weights: none
Config: configs/classification/baseline_plus/mini_imagenet/baseline-plus_resnet12_1xb64_mini-imagenet_5way-5shot.py

12
configs/classification/baseline_plus/tiered_imagenet/baseline_plus_tiered_imagenet_5way_1shot_84x84_aug.py → configs/classification/baseline_plus/tiered_imagenet/baseline-plus_conv4_1xb64_tiered-imagenet_5way-1shot.py
View File

@ -1,5 +1,5 @@
_base_ = [
'../../_base_/meta_test/tiered_imagenet_meta_test_5way_1shot_84x84.py',
'../../_base_/meta_test/tiered-imagenet_meta-test_5way-1shot.py',
'../../_base_/runtime/epoch_based_runtime.py',
'../../_base_/schedules/sgd_100epoch.py'
]
@ -25,7 +25,7 @@ meta_finetune_cfg = dict(
data = dict(
samples_per_gpu=64,
workers_per_gpu=8,
workers_per_gpu=4,
train=dict(
type='TieredImageNetDataset',
data_prefix='data/tiered_imagenet',
@ -39,4 +39,10 @@ data = dict(
meta_test_cfg=dict(
support=dict(
batch_size=4, drop_last=True, train=meta_finetune_cfg))))
pin_memory = True
model = dict(
type='BaselinePlusClassifier',
backbone=dict(type='Conv4'),
head=dict(type='CosineDistanceHead', num_classes=351, in_channels=1600),
meta_test_head=dict(
type='CosineDistanceHead', num_classes=5, in_channels=1600))

12
configs/classification/baseline_plus/tiered_imagenet/baseline_plus_tiered_imagenet_5way_5shot_84x84_aug.py → configs/classification/baseline_plus/tiered_imagenet/baseline-plus_conv4_1xb64_tiered-imagenet_5way-5shot.py
View File

@ -1,5 +1,5 @@
_base_ = [
'../../_base_/meta_test/tiered_imagenet_meta_test_5way_5shot_84x84.py',
'../../_base_/meta_test/tiered-imagenet_meta-test_5way-5shot.py',
'../../_base_/runtime/epoch_based_runtime.py',
'../../_base_/schedules/sgd_200epoch.py'
]
@ -25,7 +25,7 @@ meta_finetune_cfg = dict(
data = dict(
samples_per_gpu=64,
workers_per_gpu=8,
workers_per_gpu=4,
train=dict(
type='TieredImageNetDataset',
data_prefix='data/tiered_imagenet',
@ -39,4 +39,10 @@ data = dict(
meta_test_cfg=dict(
support=dict(
batch_size=4, drop_last=True, train=meta_finetune_cfg))))
pin_memory = True
model = dict(
type='BaselinePlusClassifier',
backbone=dict(type='Conv4'),
head=dict(type='CosineDistanceHead', num_classes=351, in_channels=1600),
meta_test_head=dict(
type='CosineDistanceHead', num_classes=5, in_channels=1600))

55
configs/classification/baseline_plus/tiered_imagenet/baseline-plus_resnet12_1xb64_tiered-imagenet_5way-1shot.py 100644
View File

@ -0,0 +1,55 @@
_base_ = [
'../../_base_/meta_test/tiered-imagenet_meta-test_5way-1shot.py',
'../../_base_/runtime/epoch_based_runtime.py',
'../../_base_/schedules/sgd_100epoch.py'
]
img_size = 84
img_norm_cfg = dict(
mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='RandomResizedCrop', size=img_size),
dict(type='RandomFlip', flip_prob=0.5, direction='horizontal'),
dict(type='ColorJitter', brightness=0.4, contrast=0.4, saturation=0.4),
dict(type='Normalize', **img_norm_cfg),
dict(type='ImageToTensor', keys=['img']),
dict(type='ToTensor', keys=['gt_label']),
dict(type='Collect', keys=['img', 'gt_label'])
]
meta_finetune_cfg = dict(
num_steps=150,
optimizer=dict(
type='SGD', lr=0.01, momentum=0.9, dampening=0.9, weight_decay=0.001))
data = dict(
samples_per_gpu=64,
workers_per_gpu=4,
train=dict(
type='TieredImageNetDataset',
data_prefix='data/tiered_imagenet',
subset='train',
pipeline=train_pipeline),
val=dict(
meta_test_cfg=dict(
support=dict(
batch_size=4, drop_last=True, train=meta_finetune_cfg))),
test=dict(
meta_test_cfg=dict(
support=dict(
batch_size=4, drop_last=True, train=meta_finetune_cfg))))
model = dict(
type='BaselinePlusClassifier',
backbone=dict(type='ResNet12'),
head=dict(
type='CosineDistanceHead',
num_classes=351,
in_channels=640,
temperature=10.0),
meta_test_head=dict(
type='CosineDistanceHead',
num_classes=5,
in_channels=640,
temperature=5.0))

55
configs/classification/baseline_plus/tiered_imagenet/baseline-plus_resnet12_1xb64_tiered-imagenet_5way-5shot.py 100644
View File

@ -0,0 +1,55 @@
_base_ = [
'../../_base_/meta_test/tiered-imagenet_meta-test_5way-5shot.py',
'../../_base_/runtime/epoch_based_runtime.py',
'../../_base_/schedules/sgd_200epoch.py'
]
img_size = 84
img_norm_cfg = dict(
mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='RandomResizedCrop', size=img_size),
dict(type='RandomFlip', flip_prob=0.5, direction='horizontal'),
dict(type='ColorJitter', brightness=0.4, contrast=0.4, saturation=0.4),
dict(type='Normalize', **img_norm_cfg),
dict(type='ImageToTensor', keys=['img']),
dict(type='ToTensor', keys=['gt_label']),
dict(type='Collect', keys=['img', 'gt_label'])
]
meta_finetune_cfg = dict(
num_steps=600,
optimizer=dict(
type='SGD', lr=0.01, momentum=0.9, dampening=0.9, weight_decay=0.001))
data = dict(
samples_per_gpu=64,
workers_per_gpu=4,
train=dict(
type='TieredImageNetDataset',
data_prefix='data/tiered_imagenet',
subset='train',
pipeline=train_pipeline),
val=dict(
meta_test_cfg=dict(
support=dict(
batch_size=4, drop_last=True, train=meta_finetune_cfg))),
test=dict(
meta_test_cfg=dict(
support=dict(
batch_size=4, drop_last=True, train=meta_finetune_cfg))))
model = dict(
type='BaselinePlusClassifier',
backbone=dict(type='ResNet12'),
head=dict(
type='CosineDistanceHead',
num_classes=351,
in_channels=640,
temperature=10.0),
meta_test_head=dict(
type='CosineDistanceHead',
num_classes=5,
in_channels=640,
temperature=5.0))

8
configs/classification/baseline_plus/tiered_imagenet/baseline_plus_conv4_tiered_imagenet_5way_1shot.py
View File

@ -1,8 +0,0 @@
_base_ = ['baseline_pp_tiered_imagenet_5way_1shot_84x84_aug.py']
model = dict(
type='BaselinePlusClassifier',
backbone=dict(type='Conv4'),
head=dict(type='CosineDistanceHead', num_classes=351, in_channels=1600),
meta_test_head=dict(
type='CosineDistanceHead', num_classes=5, in_channels=1600))

8
configs/classification/baseline_plus/tiered_imagenet/baseline_plus_conv4_tiered_imagenet_5way_5shot.py
View File

@ -1,8 +0,0 @@
_base_ = ['baseline_pp_tiered_imagenet_5way_5shot_84x84_aug.py']
model = dict(
type='BaselinePlusClassifier',
backbone=dict(type='Conv4'),
head=dict(type='CosineDistanceHead', num_classes=351, in_channels=1600),
meta_test_head=dict(
type='CosineDistanceHead', num_classes=5, in_channels=1600))

15
configs/classification/baseline_plus/tiered_imagenet/baseline_plus_resnet12_tiered_imagenet_5way_1shot.py
View File

@ -1,15 +0,0 @@
_base_ = ['baseline_pp_tiered_imagenet_5way_1shot_84x84_aug.py']
model = dict(
type='BaselinePlusClassifier',
backbone=dict(type='ResNet12'),
head=dict(
type='CosineDistanceHead',
num_classes=351,
in_channels=640,
temperature=10.0),
meta_test_head=dict(
type='CosineDistanceHead',
num_classes=5,
in_channels=640,
temperature=5.0))

15
configs/classification/baseline_plus/tiered_imagenet/baseline_plus_resnet12_tiered_imagenet_5way_5shot.py
View File

@ -1,15 +0,0 @@
_base_ = ['baseline_pp_tiered_imagenet_5way_5shot_84x84_aug.py']
model = dict(
type='BaselinePlusClassifier',
backbone=dict(type='ResNet12'),
head=dict(
type='CosineDistanceHead',
num_classes=351,
in_channels=640,
temperature=10.0),
meta_test_head=dict(
type='CosineDistanceHead',
num_classes=5,
in_channels=640,
temperature=5.0))

64
configs/classification/baseline_plus/tiered_imagenet/metafile.yml 100644
View File

@ -0,0 +1,64 @@
Collections:
- Name: Baseline_Plus
Metadata:
Training Data: Tiered-ImageNet
Training Techniques:
- SGD with Momentum
- Weight Decay
Training Resources: 1x V100 GPUs
Paper: https://arxiv.org/abs/1904.04232
README: configs/classification/baseline_plus/README.md
Models:
- Name: baseline-plus_conv4_1xb64_tiered-imagenet_5way-1shot
Metadata:
Training Data: Tiered-ImageNet
Epochs: 100
Batch Size: 64
In Collection: Baseline_Plus
Results:
- Task: Few Shot Image Classification
Dataset: Tiered-ImageNet
Metrics:
Accuracy: none
Weights: none
Config: configs/classification/baseline_plus/tiered_imagenet/baseline-plus_conv4_1xb64_tiered-imagenet_5way-1shot.py
- Name: baseline-plus_conv4_1xb64_tiered-imagenet_5way-5shot
Metadata:
Training Data: Tiered-ImageNet
Epochs: 100
Batch Size: 64
In Collection: Baseline_Plus
Results:
- Task: Few Shot Image Classification
Dataset: Tiered-ImageNet
Metrics:
Accuracy: none
Weights: none
Config: configs/classification/baseline_plus/tiered_imagenet/baseline-plus_conv4_1xb64_tiered-imagenet_5way-5shot.py
- Name: baseline-plus_resnet12_1xb64_tiered-imagenet_5way-1shot
Metadata:
Training Data: Tiered-ImageNet
Epochs: 100
Batch Size: 64
In Collection: Baseline_Plus
Results:
- Task: Few Shot Image Classification
Dataset: Tiered-ImageNet
Metrics:
Accuracy: none
Weights: none
Config: configs/classification/baseline_plus/tiered_imagenet/baseline-plus_resnet12_1xb64_tiered-imagenet_5way-1shot.py
- Name: baseline-plus_resnet12_1xb64_tiered-imagenet_5way-5shot
Metadata:
Training Data: Tiered-ImageNet
Epochs: 100
Batch Size: 64
In Collection: Baseline_Plus
Results:
- Task: Few Shot Image Classification
Dataset: Tiered-ImageNet
Metrics:
Accuracy: none
Weights: none
Config: configs/classification/baseline_plus/tiered_imagenet/baseline-plus_resnet12_1xb64_tiered-imagenet_5way-5shot.py

66
configs/classification/maml/README.md 100644
View File

@ -0,0 +1,66 @@
<!-- [ALGORITHM] -->
# <summary><a href="https://arxiv.org/abs/1703.03400"> MAML (ICML'2017)</a></summary>
```bibtex
@inproceedings{FinnAL17,
title={Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks},
author={Chelsea Finn and Pieter Abbeel and Sergey Levine},
booktitle={Proceedings of the 34th International Conference on Machine Learning},
year={2017}
}
```
## How to Reproduce MAML
It consists of two steps:
- **Step1: Base training**
- use all the images of base classes to train a base model.
- use validation set to select the best model.
- **Step2: Meta Testing**:
- use best model from step1.
### An example of CUB dataset with Conv4
```bash
# base training
python ./tools/classification/train.py \
configs/classification/maml/cub/maml_conv4_1xb105_cub_5way-1shot.py
# meta testing
python ./tools/classification/test.py \
configs/classification/maml/cub/maml_conv4_1xb105_cub_5way-1shot.py \
work_dir/maml_conv4_1xb105_cub_5way-1shot/best_accuracy_mean.pth
```
**Note**:
- All the result are trained with single gpu.
- The base training of 1 shot and 5 shot use same training setting,
but different validation setting.
## Results on CUB dataset of 1000 episodes
| Arch | Input Size | Batch Size | way | shot | mean Acc | std | ckpt | log |
| :-------------- | :-----------: | :------: | :------: | :------: | :------: | :------: |:------: |:------: |
| [conv4](/configs/classification/maml/cub/maml_conv4_1xb105_cub_5way-1shot.py) | 84x84 | 64 | 5 | 1 | - | - | [ckpt]() | [log]() |
| [conv4](/configs/classification/maml/cub/maml_conv4_1xb105_cub_5way-5shot.py) | 84x84 | 64 | 5 | 5 | - | - | [ckpt]() | [log]() |
| [resnet12](/configs/classification/maml/cub/maml_resnet12_1xb105_cub_5way-1shot.py) | 84x84 | 64 | 5 | 1 | - | - | [ckpt]() | [log]() |
| [resnet12](/configs/classification/maml/cub/maml_resnet12_1xb105_cub_5way-5shot.py) | 84x84 | 64 | 5 | 5 | - | - | [ckpt]() | [log]() |
## Results on Mini-ImageNet dataset of 1000 episodes
| Arch | Input Size | Batch Size | way | shot | mean Acc | std | ckpt | log |
| :-------------- | :-----------: | :------: | :------: | :------: | :------: | :------: |:------: |:------: |
| [conv4](/configs/classification/maml/mini_imagenet/maml_conv4_1xb105_mini-imagenet_5way-1shot.py) | 84x84 | 64 | 5 | 1 | - | - | [ckpt]() | [log]() |
| [conv4](/configs/classification/maml/mini_imagenet/maml_conv4_1xb105_mini-imagenet_5way-5shot.py) | 84x84 | 64 | 5 | 5 | - | - | [ckpt]() | [log]() |
| [resnet12](/configs/classification/maml/mini_imagenet/maml_resnet12_1xb105_mini-imagenet_5way-1shot.py) | 84x84 | 64 | 5 | 1 | - | - | [ckpt]() | [log]() |
| [resnet12](/configs/classification/maml/mini_imagenet/maml_resnet12_1xb105_mini-imagenet_5way-5shot.py) | 84x84 | 64 | 5 | 5 | - | - | [ckpt]() | [log]() |
## Results on Tiered-ImageNet of 1000 episodes
| Arch | Input Size | Batch Size | way | shot | mean Acc | std | ckpt | log |
| :-------------- | :-----------: | :------: | :------: | :------: | :------: | :------: |:------: |:------: |
| [conv4](/configs/classification/maml/tiered_imagenet/maml_conv4_1xb105_tiered-imagenet_5way-1shot.py) | 84x84 | 64 | 5 | 1 | - | - | [ckpt]() | [log]() |
| [conv4](/configs/classification/maml/tiered_imagenet/maml_conv4_1xb105_tiered-imagenet_5way-5shot.py) | 84x84 | 64 | 5 | 5 | - | - | [ckpt]() | [log]() |
| [resnet12](/configs/classification/maml/tiered_imagenet/maml_resnet12_1xb105_tiered-imagenet_5way-1shot.py) | 84x84 | 64 | 5 | 1 | - | - | [ckpt]() | [log]() |
| [resnet12](/configs/classification/maml/tiered_imagenet/maml_resnet12_1xb105_tiered-imagenet_5way-5shot.py) | 84x84 | 64 | 5 | 5 | - | - | [ckpt]() | [log]() |

14
configs/classification/maml/cub/maml_cub_5way_1shot_84x84_aug.py → configs/classification/maml/cub/maml_conv4_1xb105_cub_5way-1shot.py
View File

@ -1,7 +1,7 @@
_base_ = [
'../../_base_/meta_test/cub_meta_test_5way_1shot_84x84.py',
'../../_base_/meta_test/cub_meta-test_5way-1shot.py',
'../../_base_/runtime/iter_based_runtime.py',
'../../_base_/schedules/adam_100000iter.py'
'../../_base_/schedules/adam_100k_iter.py'
]
img_size = 84
@ -20,7 +20,7 @@ train_pipeline = [
data = dict(
samples_per_gpu=1,
workers_per_gpu=6,
workers_per_gpu=4,
train=dict(
type='EpisodicDataset',
num_episodes=100000,
@ -39,6 +39,14 @@ data = dict(
meta_test_cfg=dict(
fast_test=False, support=dict(batch_size=5, num_inner_steps=5))))
model = dict(
type='MAMLClassifier',
num_inner_steps=2,
inner_lr=0.01,
first_order=False,
backbone=dict(type='Conv4'),
head=dict(type='LinearHead', num_classes=5, in_channels=1600))
optimizer_config = dict(
type='GradientCumulativeOptimizerHook', cumulative_iters=8, grad_clip=None)
optimizer = dict(type='Adam', lr=0.001)

14
configs/classification/maml/cub/maml_cub_5way_5shot_84x84_aug.py → configs/classification/maml/cub/maml_conv4_1xb105_cub_5way-5shot.py
View File

@ -1,7 +1,7 @@
_base_ = [
'../../_base_/meta_test/cub_meta_test_5way_5shot_84x84.py',
'../../_base_/meta_test/cub_meta-test_5way-5shot.py',
'../../_base_/runtime/iter_based_runtime.py',
'../../_base_/schedules/adam_100000iter.py'
'../../_base_/schedules/adam_100k_iter.py'
]
img_size = 84
@ -20,7 +20,7 @@ train_pipeline = [
data = dict(
samples_per_gpu=1,
workers_per_gpu=6,
workers_per_gpu=4,
train=dict(
type='EpisodicDataset',
num_episodes=100000,
@ -39,6 +39,14 @@ data = dict(
meta_test_cfg=dict(
fast_test=False, support=dict(batch_size=25, num_inner_steps=5))))
model = dict(
type='MAMLClassifier',
num_inner_steps=2,
inner_lr=0.01,
first_order=False,
backbone=dict(type='Conv4'),
head=dict(type='LinearHead', num_classes=5, in_channels=1600))
optimizer_config = dict(
type='GradientCumulativeOptimizerHook', cumulative_iters=8, grad_clip=None)
optimizer = dict(type='Adam', lr=0.001)

10
configs/classification/maml/cub/maml_conv4_cub_5way_1shot.py
View File

@ -1,10 +0,0 @@
_base_ = [
'maml_cub_5way_1shot_84x84_aug.py',
]
model = dict(
type='MAMLClassifier',
num_inner_steps=2,
inner_lr=0.01,
first_order=False,
backbone=dict(type='Conv4'),
head=dict(type='LinearHead', num_classes=5, in_channels=1600))

10
configs/classification/maml/cub/maml_conv4_cub_5way_5shot.py
View File

@ -1,10 +0,0 @@
_base_ = [
'maml_cub_5way_5shot_84x84_aug.py',
]
model = dict(
type='MAMLClassifier',
num_inner_steps=2,
inner_lr=0.01,
first_order=False,
backbone=dict(type='Conv4'),
head=dict(type='LinearHead', num_classes=5, in_channels=1600))

52
configs/classification/maml/cub/maml_resnet12_1xb105_cub_5way-1shot.py 100644
View File

@ -0,0 +1,52 @@
_base_ = [
'../../_base_/meta_test/cub_meta-test_5way-1shot.py',
'../../_base_/runtime/iter_based_runtime.py',
'../../_base_/schedules/adam_100k_iter.py'
]
img_size = 84
img_norm_cfg = dict(
mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='RandomResizedCrop', size=img_size),
dict(type='RandomFlip', flip_prob=0.5, direction='horizontal'),
dict(type='ColorJitter', brightness=0.4, contrast=0.4, saturation=0.4),
dict(type='Normalize', **img_norm_cfg),
dict(type='ImageToTensor', keys=['img']),
dict(type='ToTensor', keys=['gt_label']),
dict(type='Collect', keys=['img', 'gt_label'])
]
data = dict(
samples_per_gpu=1,
workers_per_gpu=4,
train=dict(
type='EpisodicDataset',
num_episodes=100000,
num_ways=5,
num_shots=5,
num_queries=16,
dataset=dict(
type='CUBDataset',
data_prefix='data/CUB_200_2011',
subset='train',
pipeline=train_pipeline)),
val=dict(
meta_test_cfg=dict(
fast_test=False, support=dict(batch_size=5, num_inner_steps=5))),
test=dict(
meta_test_cfg=dict(
fast_test=False, support=dict(batch_size=5, num_inner_steps=5))))
model = dict(
type='MAMLClassifier',
num_inner_steps=2,
inner_lr=0.01,
first_order=False,
backbone=dict(type='ResNet12'),
head=dict(type='LinearHead', num_classes=5, in_channels=640))
optimizer_config = dict(
type='GradientCumulativeOptimizerHook', cumulative_iters=8, grad_clip=None)
optimizer = dict(type='Adam', lr=0.001)

52
configs/classification/maml/cub/maml_resnet12_1xb105_cub_5way-5shot.py 100644
View File

@ -0,0 +1,52 @@
_base_ = [
'../../_base_/meta_test/cub_meta-test_5way-5shot.py',
'../../_base_/runtime/iter_based_runtime.py',
'../../_base_/schedules/adam_100k_iter.py'
]
img_size = 84
img_norm_cfg = dict(
mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='RandomResizedCrop', size=img_size),
dict(type='RandomFlip', flip_prob=0.5, direction='horizontal'),
dict(type='ColorJitter', brightness=0.4, contrast=0.4, saturation=0.4),
dict(type='Normalize', **img_norm_cfg),
dict(type='ImageToTensor', keys=['img']),
dict(type='ToTensor', keys=['gt_label']),
dict(type='Collect', keys=['img', 'gt_label'])
]
data = dict(
samples_per_gpu=1,
workers_per_gpu=4,
train=dict(
type='EpisodicDataset',
num_episodes=100000,
num_ways=5,
num_shots=5,
num_queries=16,
dataset=dict(
type='CUBDataset',
data_prefix='data/CUB_200_2011',
subset='train',
pipeline=train_pipeline)),
val=dict(
meta_test_cfg=dict(
fast_test=False, support=dict(batch_size=25, num_inner_steps=5))),
test=dict(
meta_test_cfg=dict(
fast_test=False, support=dict(batch_size=25, num_inner_steps=5))))
model = dict(
type='MAMLClassifier',
num_inner_steps=2,
inner_lr=0.01,
first_order=False,
backbone=dict(type='ResNet12'),
head=dict(type='LinearHead', num_classes=5, in_channels=640))
optimizer_config = dict(
type='GradientCumulativeOptimizerHook', cumulative_iters=8, grad_clip=None)
optimizer = dict(type='Adam', lr=0.001)

10
configs/classification/maml/cub/maml_resnet12_cub_5way_1shot.py
View File

@ -1,10 +0,0 @@
_base_ = [
'maml_cub_5way_1shot_84x84_aug.py',
]
model = dict(
type='MAMLClassifier',
num_inner_steps=2,
inner_lr=0.01,
first_order=False,
backbone=dict(type='ResNet12'),
head=dict(type='LinearHead', num_classes=5, in_channels=640))

10
configs/classification/maml/cub/maml_resnet12_cub_5way_5shot.py
View File

@ -1,10 +0,0 @@
_base_ = [
'maml_cub_5way_5shot_84x84_aug.py',
]
model = dict(
type='MAMLClassifier',
num_inner_steps=2,
inner_lr=0.01,
first_order=False,
backbone=dict(type='ResNet12'),
head=dict(type='LinearHead', num_classes=5, in_channels=640))

59
configs/classification/maml/cub/metafile.yml 100644
View File

@ -0,0 +1,59 @@
Collections:
- Name: MAML
Metadata:
Training Data: CUB
Training Techniques:
- Adam
Training Resources: 1x V100 GPUs
Paper: https://arxiv.org/abs/1703.03400
README: configs/classification/maml/README.md
Models:
- Name: maml_conv4_1xb105_cub_5way-1shot
Metadata:
Training Data: CUB
Batch Size: 105
In Collection: MAML
Results:
- Task: Few Shot Image Classification
Dataset: CUB
Metrics:
Accuracy: none
Weights: none
Config: configs/classification/maml/cub/maml_conv4_1xb105_cub_5way-1shot.py
- Name: maml_conv4_1xb105_cub_5way-5shot
Metadata:
Training Data: CUB
Batch Size: 105
In Collection: MAML
Results:
- Task: Few Shot Image Classification
Dataset: CUB
Metrics:
Accuracy: none
Weights: none
Config: configs/classification/maml/cub/maml_conv4_1xb105_cub_5way-5shot.py
- Name: maml_resnet12_1xb105_cub_5way-1shot
Metadata:
Training Data: CUB
Batch Size: 105
In Collection: MAML
Results:
- Task: Few Shot Image Classification
Dataset: CUB
Metrics:
Accuracy: none
Weights: none
Config: configs/classification/maml/cub/maml_resnet12_1xb105_cub_5way-1shot.py
- Name: maml_resnet12_1xb105_cub_5way-5shot
Metadata:
Training Data: CUB
Batch Size: 105
In Collection: MAML
Results:
- Task: Few Shot Image Classification
Dataset: CUB
Metrics:
Accuracy: none
Weights: none
Config: configs/classification/maml/cub/maml_resnet12_1xb105_cub_5way-5shot.py

14
configs/classification/maml/mini_imagenet/maml_mini_imagenet_5way_1shot_84x84_aug.py → configs/classification/maml/mini_imagenet/maml_conv4_1xb105_mini-imagenet_5way-1shot.py
View File

@ -1,7 +1,7 @@
_base_ = [
'../../_base_/meta_test/mini_imagenet_meta_test_5way_1shot_84x84.py',
'../../_base_/meta_test/mini-imagenet_meta-test_5way-1shot.py',
'../../_base_/runtime/iter_based_runtime.py',
'../../_base_/schedules/adam_100000iter.py'
'../../_base_/schedules/adam_100k_iter.py'
]
img_size = 84
@ -20,7 +20,7 @@ train_pipeline = [
data = dict(
samples_per_gpu=1,
workers_per_gpu=6,
workers_per_gpu=4,
train=dict(
type='EpisodicDataset',
num_episodes=100000,
@ -39,6 +39,14 @@ data = dict(
meta_test_cfg=dict(
fast_test=False, support=dict(batch_size=5, num_inner_steps=5))))
model = dict(
type='MAMLClassifier',
num_inner_steps=2,
inner_lr=0.01,
first_order=False,
backbone=dict(type='Conv4'),
head=dict(type='LinearHead', num_classes=5, in_channels=1600))
optimizer_config = dict(
type='GradientCumulativeOptimizerHook', cumulative_iters=8, grad_clip=None)
optimizer = dict(type='Adam', lr=0.001)

14
configs/classification/maml/mini_imagenet/maml_mini_imagenet_5way_5shot_84x84_aug.py → configs/classification/maml/mini_imagenet/maml_conv4_1xb105_mini-imagenet_5way-5shot.py
View File

@ -1,7 +1,7 @@
_base_ = [
'../../_base_/meta_test/mini_imagenet_meta_test_5way_5shot_84x84.py',
'../../_base_/meta_test/mini-imagenet_meta-test_5way-5shot.py',
'../../_base_/runtime/iter_based_runtime.py',
'../../_base_/schedules/adam_100000iter.py'
'../../_base_/schedules/adam_100k_iter.py'
]
img_size = 84
@ -20,7 +20,7 @@ train_pipeline = [
data = dict(
samples_per_gpu=1,
workers_per_gpu=6,
workers_per_gpu=4,
train=dict(
type='EpisodicDataset',
num_episodes=100000,
@ -39,6 +39,14 @@ data = dict(
meta_test_cfg=dict(
fast_test=False, support=dict(batch_size=25, num_inner_steps=5))))
model = dict(
type='MAMLClassifier',
num_inner_steps=2,
inner_lr=0.01,
first_order=False,
backbone=dict(type='Conv4'),
head=dict(type='LinearHead', num_classes=5, in_channels=1600))
optimizer_config = dict(
type='GradientCumulativeOptimizerHook', cumulative_iters=8, grad_clip=None)
optimizer = dict(type='Adam', lr=0.001)

10
configs/classification/maml/mini_imagenet/maml_conv4_mini_imagenet_5way_1shot.py
View File

@ -1,10 +0,0 @@
_base_ = [
'maml_mini_imagenet_5way_1shot_84x84_aug.py',
]
model = dict(
type='MAMLClassifier',
num_inner_steps=2,
inner_lr=0.01,
first_order=False,
backbone=dict(type='Conv4'),
head=dict(type='LinearHead', num_classes=5, in_channels=1600))

10
configs/classification/maml/mini_imagenet/maml_conv4_mini_imagenet_5way_5shot.py
View File

@ -1,10 +0,0 @@
_base_ = [
'maml_mini_imagenet_5way_5shot_84x84_aug.py',
]
model = dict(
type='MAMLClassifier',
num_inner_steps=2,
inner_lr=0.01,
first_order=False,
backbone=dict(type='Conv4'),
head=dict(type='LinearHead', num_classes=5, in_channels=1600))

52
configs/classification/maml/mini_imagenet/maml_resnet12_1xb105_mini-imagenet_5way-1shot.py 100644
View File

@ -0,0 +1,52 @@
_base_ = [
'../../_base_/meta_test/mini-imagenet_meta-test_5way-1shot.py',
'../../_base_/runtime/iter_based_runtime.py',
'../../_base_/schedules/adam_100k_iter.py'
]
img_size = 84
img_norm_cfg = dict(
mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='RandomResizedCrop', size=img_size),
dict(type='RandomFlip', flip_prob=0.5, direction='horizontal'),
dict(type='ColorJitter', brightness=0.4, contrast=0.4, saturation=0.4),
dict(type='Normalize', **img_norm_cfg),
dict(type='ImageToTensor', keys=['img']),
dict(type='ToTensor', keys=['gt_label']),
dict(type='Collect', keys=['img', 'gt_label'])
]
data = dict(
samples_per_gpu=1,
workers_per_gpu=4,
train=dict(
type='EpisodicDataset',
num_episodes=100000,
num_ways=5,
num_shots=5,
num_queries=16,
dataset=dict(
type='MiniImageNetDataset',
data_prefix='data/mini_imagenet',
subset='train',
pipeline=train_pipeline)),
val=dict(
meta_test_cfg=dict(
fast_test=False, support=dict(batch_size=5, num_inner_steps=5))),
test=dict(
meta_test_cfg=dict(
fast_test=False, support=dict(batch_size=5, num_inner_steps=5))))
model = dict(
type='MAMLClassifier',
num_inner_steps=2,
inner_lr=0.01,
first_order=False,
backbone=dict(type='ResNet12'),
head=dict(type='LinearHead', num_classes=5, in_channels=640))
optimizer = dict(type='Adam', lr=0.001)
optimizer_config = dict(
type='GradientCumulativeOptimizerHook', cumulative_iters=8, grad_clip=None)

52
configs/classification/maml/mini_imagenet/maml_resnet12_1xb105_mini-imagenet_5way-5shot.py 100644
View File

@ -0,0 +1,52 @@
_base_ = [
'../../_base_/meta_test/mini-imagenet_meta-test_5way-5shot.py',
'../../_base_/runtime/iter_based_runtime.py',
'../../_base_/schedules/adam_100k_iter.py'
]
img_size = 84
img_norm_cfg = dict(
mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='RandomResizedCrop', size=img_size),
dict(type='RandomFlip', flip_prob=0.5, direction='horizontal'),
dict(type='ColorJitter', brightness=0.4, contrast=0.4, saturation=0.4),
dict(type='Normalize', **img_norm_cfg),
dict(type='ImageToTensor', keys=['img']),
dict(type='ToTensor', keys=['gt_label']),
dict(type='Collect', keys=['img', 'gt_label'])
]
data = dict(
samples_per_gpu=1,
workers_per_gpu=4,
train=dict(
type='EpisodicDataset',
num_episodes=100000,
num_ways=5,
num_shots=5,
num_queries=16,
dataset=dict(
type='MiniImageNetDataset',
data_prefix='data/mini_imagenet',
subset='train',
pipeline=train_pipeline)),
val=dict(
meta_test_cfg=dict(
fast_test=False, support=dict(batch_size=25, num_inner_steps=5))),
test=dict(
meta_test_cfg=dict(
fast_test=False, support=dict(batch_size=25, num_inner_steps=5))))
model = dict(
type='MAMLClassifier',
num_inner_steps=2,
inner_lr=0.01,
first_order=False,
backbone=dict(type='ResNet12'),
head=dict(type='LinearHead', num_classes=5, in_channels=640))
optimizer_config = dict(
type='GradientCumulativeOptimizerHook', cumulative_iters=8, grad_clip=None)
optimizer = dict(type='Adam', lr=0.001)

10
configs/classification/maml/mini_imagenet/maml_resnet12_mini_imagenet_5way_1shot.py
View File

@ -1,10 +0,0 @@
_base_ = [
'maml_mini_imagenet_5way_1shot_84x84_aug.py',
]
model = dict(
type='MAMLClassifier',
num_inner_steps=2,
inner_lr=0.01,
first_order=False,
backbone=dict(type='ResNet12'),
head=dict(type='LinearHead', num_classes=5, in_channels=640))

10
configs/classification/maml/mini_imagenet/maml_resnet12_mini_imagenet_5way_5shot.py
View File

@ -1,10 +0,0 @@
_base_ = [
'maml_mini_imagenet_5way_5shot_84x84_aug.py',
]
model = dict(
type='MAMLClassifier',
num_inner_steps=2,
inner_lr=0.01,
first_order=False,
backbone=dict(type='ResNet12'),
head=dict(type='LinearHead', num_classes=5, in_channels=640))

59
configs/classification/maml/mini_imagenet/metafile.yml 100644
View File

@ -0,0 +1,59 @@
Collections:
- Name: MAML
Metadata:
Training Data: Mini-ImageNet
Training Techniques:
- Adam
Training Resources: 1x V100 GPUs
Paper: https://arxiv.org/abs/1703.03400
README: configs/classification/maml/README.md
Models:
- Name: maml_conv4_1xb105_mini-imagenet_5way-1shot
Metadata:
Training Data: Mini-ImageNet
Batch Size: 105
In Collection: MAML
Results:
- Task: Few Shot Image Classification
Dataset: Mini-ImageNet
Metrics:
Accuracy: none
Weights: none
Config: configs/classification/maml/mini_imagenet/maml_conv4_1xb105_mini-imagenet_5way-1shot.py
- Name: maml_conv4_1xb105_mini-imagenet_5way-5shot
Metadata:
Training Data: Mini-ImageNet
Batch Size: 105
In Collection: MAML
Results:
- Task: Few Shot Image Classification
Dataset: Mini-ImageNet
Metrics:
Accuracy: none
Weights: none
Config: configs/classification/maml/mini_imagenet/maml_conv4_1xb105_mini-imagenet_5way-5shot.py
- Name: maml_resnet12_1xb105_mini-imagenet_5way-1shot
Metadata:
Training Data: Mini-ImageNet
Batch Size: 105
In Collection: MAML
Results:
- Task: Few Shot Image Classification
Dataset: Mini-ImageNet
Metrics:
Accuracy: none
Weights: none
Config: configs/classification/maml/mini_imagenet/maml_resnet12_1xb105_mini-imagenet_5way-1shot.py
- Name: maml_resnet12_1xb105_mini-imagenet_5way-5shot
Metadata:
Training Data: Mini-ImageNet
Batch Size: 105
In Collection: MAML
Results:
- Task: Few Shot Image Classification
Dataset: Mini-ImageNet
Metrics:
Accuracy: none
Weights: none
Config: configs/classification/maml/mini_imagenet/maml_resnet12_1xb105_mini-imagenet_5way-5shot.py

15
configs/classification/maml/tiered_imagenet/maml_tiered_imagenet_5way_1shot_84x84_aug.py → configs/classification/maml/tiered_imagenet/maml_conv4_1xb105_tiered-imagenet_5way-1shot.py
View File

@ -1,7 +1,7 @@
_base_ = [
'../../_base_/meta_test/tiered_imagenet_meta_test_5way_1shot_84x84.py',
'../../_base_/meta_test/tiered-imagenet_meta-test_5way-1shot.py',
'../../_base_/runtime/iter_based_runtime.py',
'../../_base_/schedules/adam_100000iter.py'
'../../_base_/schedules/adam_100k_iter.py'
]
img_size = 84
@ -20,7 +20,7 @@ train_pipeline = [
data = dict(
samples_per_gpu=1,
workers_per_gpu=6,
workers_per_gpu=4,
train=dict(
type='EpisodicDataset',
num_episodes=100000,
@ -39,7 +39,14 @@ data = dict(
meta_test_cfg=dict(
fast_test=False, support=dict(batch_size=5, num_inner_steps=5))))
model = dict(
type='MAMLClassifier',
num_inner_steps=2,
inner_lr=0.01,
first_order=False,
backbone=dict(type='Conv4'),
head=dict(type='LinearHead', num_classes=5, in_channels=1600))
optimizer_config = dict(
type='GradientCumulativeOptimizerHook', cumulative_iters=8, grad_clip=None)
optimizer = dict(type='Adam', lr=0.001)
pin_memory = True

15
configs/classification/maml/tiered_imagenet/maml_tiered_imagenet_5way_5shot_84x84_aug.py → configs/classification/maml/tiered_imagenet/maml_conv4_1xb105_tiered-imagenet_5way-5shot.py
View File

@ -1,7 +1,7 @@
_base_ = [
'../../_base_/meta_test/tiered_imagenet_meta_test_5way_5shot_84x84.py',
'../../_base_/meta_test/tiered-imagenet_meta-test_5way-5shot.py',
'../../_base_/runtime/iter_based_runtime.py',
'../../_base_/schedules/adam_100000iter.py'
'../../_base_/schedules/adam_100k_iter.py'
]
img_size = 84
@ -20,7 +20,7 @@ train_pipeline = [
data = dict(
samples_per_gpu=1,
workers_per_gpu=6,
workers_per_gpu=4,
train=dict(
type='EpisodicDataset',
num_episodes=100000,
@ -39,7 +39,14 @@ data = dict(
meta_test_cfg=dict(
fast_test=False, support=dict(batch_size=25, num_inner_steps=5))))
model = dict(
type='MAMLClassifier',
num_inner_steps=2,
inner_lr=0.01,
first_order=False,
backbone=dict(type='Conv4'),
head=dict(type='LinearHead', num_classes=5, in_channels=1600))
optimizer_config = dict(
type='GradientCumulativeOptimizerHook', cumulative_iters=8, grad_clip=None)
optimizer = dict(type='Adam', lr=0.001)
pin_memory = True

10
configs/classification/maml/tiered_imagenet/maml_conv4_tiered_imagenet_5way_1shot.py
View File

@ -1,10 +0,0 @@
_base_ = [
'maml_tiered_imagenet_5way_1shot_84x84_aug.py',
]
model = dict(
type='MAMLClassifier',
num_inner_steps=2,
inner_lr=0.01,
first_order=False,
backbone=dict(type='Conv4'),
head=dict(type='LinearHead', num_classes=5, in_channels=1600))

10
configs/classification/maml/tiered_imagenet/maml_conv4_tiered_imagenet_5way_5shot.py
View File

@ -1,10 +0,0 @@
_base_ = [
'maml_tiered_imagenet_5way_5shot_84x84_aug.py',
]
model = dict(
type='MAMLClassifier',
num_inner_steps=2,
inner_lr=0.01,
first_order=False,
backbone=dict(type='Conv4'),
head=dict(type='LinearHead', num_classes=5, in_channels=1600))

52
configs/classification/maml/tiered_imagenet/maml_resnet12_1xb105_tiered-imagenet_5way-1shot.py 100644
View File

@ -0,0 +1,52 @@
_base_ = [
'../../_base_/meta_test/tiered-imagenet_meta-test_5way-1shot.py',
'../../_base_/runtime/iter_based_runtime.py',
'../../_base_/schedules/adam_100k_iter.py'
]
img_size = 84
img_norm_cfg = dict(
mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
train_pipeline = [
dict(type='LoadImageFromBytes'),
dict(type='RandomResizedCrop', size=img_size),
dict(type='RandomFlip', flip_prob=0.5, direction='horizontal'),
dict(type='ColorJitter', brightness=0.4, contrast=0.4, saturation=0.4),
dict(type='Normalize', **img_norm_cfg),
dict(type='ImageToTensor', keys=['img']),
dict(type='ToTensor', keys=['gt_label']),
dict(type='Collect', keys=['img', 'gt_label'])
]
data = dict(
samples_per_gpu=1,
workers_per_gpu=4,
train=dict(
type='EpisodicDataset',
num_episodes=100000,
num_ways=5,
num_shots=5,
num_queries=16,
dataset=dict(
type='TieredImageNetDataset',
data_prefix='data/tiered_imagenet',
subset='train',
pipeline=train_pipeline)),
val=dict(
meta_test_cfg=dict(
fast_test=False, support=dict(batch_size=5, num_inner_steps=5))),
test=dict(
meta_test_cfg=dict(
fast_test=False, support=dict(batch_size=5, num_inner_steps=5))))
model = dict(
type='MAMLClassifier',
num_inner_steps=2,
inner_lr=0.01,
first_order=False,
backbone=dict(type='ResNet12'),
head=dict(type='LinearHead', num_classes=5, in_channels=640))
optimizer_config = dict(
type='GradientCumulativeOptimizerHook', cumulative_iters=8, grad_clip=None)
optimizer = dict(type='Adam', lr=0.001)

52
configs/classification/maml/tiered_imagenet/maml_resnet12_1xb105_tiered-imagenet_5way-5shot.py 100644
View File

@ -0,0 +1,52 @@
_base_ = [
'../../_base_/meta_test/tiered-imagenet_meta-test_5way-5shot.py',
'../../_base_/runtime/iter_based_runtime.py',
'../../_base_/schedules/adam_100k_iter.py'
]
img_size = 84
img_norm_cfg = dict(
mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
train_pipeline = [
dict(type='LoadImageFromBytes'),
dict(type='RandomResizedCrop', size=img_size),
dict(type='RandomFlip', flip_prob=0.5, direction='horizontal'),
dict(type='ColorJitter', brightness=0.4, contrast=0.4, saturation=0.4),
dict(type='Normalize', **img_norm_cfg),
dict(type='ImageToTensor', keys=['img']),
dict(type='ToTensor', keys=['gt_label']),
dict(type='Collect', keys=['img', 'gt_label'])
]
data = dict(
samples_per_gpu=1,
workers_per_gpu=4,
train=dict(
type='EpisodicDataset',
num_episodes=100000,
num_ways=5,
num_shots=5,
num_queries=16,
dataset=dict(
type='TieredImageNetDataset',
data_prefix='data/tiered_imagenet',
subset='train',
pipeline=train_pipeline)),
val=dict(
meta_test_cfg=dict(
fast_test=False, support=dict(batch_size=25, num_inner_steps=5))),
test=dict(
meta_test_cfg=dict(
fast_test=False, support=dict(batch_size=25, num_inner_steps=5))))
model = dict(
type='MAMLClassifier',
num_inner_steps=2,
inner_lr=0.01,
first_order=False,
backbone=dict(type='ResNet12'),
head=dict(type='LinearHead', num_classes=5, in_channels=640))
optimizer_config = dict(
type='GradientCumulativeOptimizerHook', cumulative_iters=8, grad_clip=None)
optimizer = dict(type='Adam', lr=0.001)

10
configs/classification/maml/tiered_imagenet/maml_resnet12_tiered_imagenet_5way_1shot.py
View File

@ -1,10 +0,0 @@
_base_ = [
'maml_tiered_imagenet_5way_1shot_84x84_aug.py',
]
model = dict(
type='MAMLClassifier',
num_inner_steps=2,
inner_lr=0.01,
first_order=False,
backbone=dict(type='ResNet12'),
head=dict(type='LinearHead', num_classes=5, in_channels=640))

10
configs/classification/maml/tiered_imagenet/maml_resnet12_tiered_imagenet_5way_5shot.py
View File

@ -1,10 +0,0 @@
_base_ = [
'maml_tiered_imagenet_5way_5shot_84x84_aug.py',
]
model = dict(
type='MAMLClassifier',
num_inner_steps=2,
inner_lr=0.01,
first_order=False,
backbone=dict(type='ResNet12'),
head=dict(type='LinearHead', num_classes=5, in_channels=640))

59
configs/classification/maml/tiered_imagenet/metafile.yml 100644
View File

@ -0,0 +1,59 @@
Collections:
- Name: MAML
Metadata:
Training Data: Tiered-ImageNet
Training Techniques:
- Adam
Training Resources: 1x V100 GPUs
Paper: https://arxiv.org/abs/1703.03400
README: configs/classification/maml/README.md
Models:
- Name: maml_conv4_1xb105_tiered-imagenet_5way-1shot
Metadata:
Training Data: Tiered-ImageNet
Batch Size: 105
In Collection: MAML
Results:
- Task: Few Shot Image Classification
Dataset: Tiered-ImageNet
Metrics:
Accuracy: none
Weights: none
Config: configs/classification/maml/tiered_imagenet/maml_conv4_1xb105_tiered-imagenet_5way-1shot.py
- Name: maml_conv4_1xb105_tiered-imagenet_5way-5shot
Metadata:
Training Data: Tiered-ImageNet
Batch Size: 105
In Collection: MAML
Results:
- Task: Few Shot Image Classification
Dataset: Tiered-ImageNet
Metrics:
Accuracy: none
Weights: none
Config: configs/classification/maml/tiered_imagenet/maml_conv4_1xb105_tiered-imagenet_5way-5shot.py
- Name: maml_resnet12_1xb105_tiered-imagenet_5way-1shot
Metadata:
Training Data: Tiered-ImageNet
Batch Size: 105
In Collection: MAML
Results:
- Task: Few Shot Image Classification
Dataset: Tiered-ImageNet
Metrics:
Accuracy: none
Weights: none
Config: configs/classification/maml/tiered_imagenet/maml_resnet12_1xb105_tiered-imagenet_5way-1shot.py
- Name: maml_resnet12_1xb105_tiered-imagenet_5way-5shot
Metadata:
Training Data: Tiered-ImageNet
Batch Size: 105
In Collection: MAML
Results:
- Task: Few Shot Image Classification
Dataset: Tiered-ImageNet
Metrics:
Accuracy: none
Weights: none
Config: configs/classification/maml/tiered_imagenet/maml_resnet12_1xb105_tiered-imagenet_5way-5shot.py

68
configs/classification/matching_net/README.md 100644
View File

@ -0,0 +1,68 @@
<!-- [ALGORITHM] -->
# <summary><a href="https://arxiv.org/abs/1606.04080"> MatchingNet (NeurIPS'2016)</a></summary>
```bibtex
@inproceedings{vinyals2016matching,
title={Matching networks for one shot learning},
author={Vinyals, Oriol and Blundell, Charles and Lillicrap, Tim and Wierstra, Daan and others},
booktitle={Advances in Neural Information Processing Systems},
pages={3630--3638},
year={2016}
}
```
## How to Reproduce MatchingNet
It consists of two steps:
- **Step1: Base training**
- use all the images of base classes to train a base model.
- use validation set to select the best model.
- **Step2: Meta Testing**:
- use best model from step1.
### An example of CUB dataset with Conv4
```bash
# base training
python ./tools/classification/train.py \
configs/classification/matching_net/cub/matching-net_conv4_1xb105_cub_5way-1shot.py
# meta testing
python ./tools/classification/test.py \
configs/classification/matching_net/cub/matching-net_conv4_1xb105_cub_5way-1shot.py \
work_dir/matching-net_conv4_1xb105_cub_5way-1shot/best_accuracy_mean.pth
```
**Note**:
- All the result are trained with single gpu.
- The base training of 1 shot and 5 shot use same training setting,
but different validation setting.
## Results on CUB dataset of 1000 episodes
| Arch | Input Size | Batch Size | way | shot | mean Acc | std | ckpt | log |
| :-------------- | :-----------: | :------: | :------: | :------: | :------: | :------: |:------: |:------: |
| [conv4](/configs/classification/matching_net/cub/matching-net_conv4_1xb105_cub_5way-1shot.py) | 84x84 | 64 | 5 | 1 | - | - | [ckpt]() | [log]() |
| [conv4](/configs/classification/matching_net/cub/matching-net_conv4_1xb105_cub_5way-5shot.py) | 84x84 | 64 | 5 | 5 | - | - | [ckpt]() | [log]() |
| [resnet12](/configs/classification/matching_net/cub/matching-net_resnet12_1xb105_cub_5way-1shot.py) | 84x84 | 64 | 5 | 1 | - | - | [ckpt]() | [log]() |
| [resnet12](/configs/classification/matching_net/cub/matching-net_resnet12_1xb105_cub_5way-5shot.py) | 84x84 | 64 | 5 | 5 | - | - | [ckpt]() | [log]() |
## Results on Mini-ImageNet dataset of 1000 episodes
| Arch | Input Size | Batch Size | way | shot | mean Acc | std | ckpt | log |
| :-------------- | :-----------: | :------: | :------: | :------: | :------: | :------: |:------: |:------: |
| [conv4](/configs/classification/matching_net/mini_imagenet/matching-net_conv4_1xb105_mini-imagenet_5way-1shot.py) | 84x84 | 64 | 5 | 1 | - | - | [ckpt]() | [log]() |
| [conv4](/configs/classification/matching_net/mini_imagenet/matching-net_conv4_1xb105_mini-imagenet_5way-5shot.py) | 84x84 | 64 | 5 | 5 | - | - | [ckpt]() | [log]() |
| [resnet12](/configs/classification/matching_net/mini_imagenet/matching-net_resnet12_1xb105_mini-imagenet_5way-1shot.py) | 84x84 | 64 | 5 | 1 | - | - | [ckpt]() | [log]() |
| [resnet12](/configs/classification/matching_net/mini_imagenet/matching-net_resnet12_1xb105_mini-imagenet_5way-5shot.py) | 84x84 | 64 | 5 | 5 | - | - | [ckpt]() | [log]() |
## Results on Tiered-ImageNet of 1000 episodes
| Arch | Input Size | Batch Size | way | shot | mean Acc | std | ckpt | log |
| :-------------- | :-----------: | :------: | :------: | :------: | :------: | :------: |:------: |:------: |
| [conv4](/configs/classification/matching_net/tiered_imagenet/matching-net_conv4_1xb105_tiered-imagenet_5way-1shot.py) | 84x84 | 64 | 5 | 1 | - | - | [ckpt]() | [log]() |
| [conv4](/configs/classification/matching_net/tiered_imagenet/matching-net_conv4_1xb105_tiered-imagenet_5way-5shot.py) | 84x84 | 64 | 5 | 5 | - | - | [ckpt]() | [log]() |
| [resnet12](/configs/classification/matching_net/tiered_imagenet/matching-net_resnet12_1xb105_tiered-imagenet_5way-1shot.py) | 84x84 | 64 | 5 | 1 | - | - | [ckpt]() | [log]() |
| [resnet12](/configs/classification/matching_net/tiered_imagenet/matching-net_resnet12_1xb105_tiered-imagenet_5way-5shot.py) | 84x84 | 64 | 5 | 5 | - | - | [ckpt]() | [log]() |

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