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

[Feature] Support P6 YOLOv5 (#168) 

* add yolov5s config

* support train

* fix some comment

* add config

* add link

* UPDATE

* UPDATE

* remove
pull/249/head
Haian Huang(深度眸) 2022-10-31 14:11:27 +08:00 committed by Haian Huang(深度眸)
parent 14d3f294ef
commit b95bf63916
25 changed files with 370 additions and 52 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
README.md
View File

@ -86,7 +86,7 @@ conda activate open-mmlab
pip install openmim
mim install "mmengine>=0.2.0"
mim install "mmcv>=2.0.0rc1,<2.1.0"
mim install "mmdet>=3.0.0rc1,<3.1.0"
mim install "mmdet>=3.0.0rc2,<3.1.0"
git clone https://github.com/open-mmlab/mmyolo.git
cd mmyolo
# Install albumentations

2
README_zh-CN.md
View File

@ -87,7 +87,7 @@ conda activate open-mmlab
pip install openmim
mim install "mmengine>=0.2.0"
mim install "mmcv>=2.0.0rc1,<2.1.0"
mim install "mmdet>=3.0.0rc1,<3.1.0"
mim install "mmdet>=3.0.0rc2,<3.1.0"
git clone https://github.com/open-mmlab/mmyolo.git
cd mmyolo
# Install albumentations

1
configs/rtmdet/rtmdet_l_syncbn_8xb32-300e_coco.py
View File

@ -218,6 +218,7 @@ custom_hooks = [
ema_type='ExpMomentumEMA',
momentum=0.0002,
update_buffers=True,
strict_load=False,
priority=49),
dict(
type='mmdet.PipelineSwitchHook',

1
configs/rtmdet/rtmdet_s_syncbn_8xb32-300e_coco.py
View File

@ -70,6 +70,7 @@ custom_hooks = [
ema_type='ExpMomentumEMA',
momentum=0.0002,
update_buffers=True,
strict_load=False,
priority=49),
dict(
type='mmdet.PipelineSwitchHook',

21
configs/yolov5/README.md
View File

@ -10,20 +10,25 @@ YOLOv5 is a family of object detection architectures and models pretrained on th
### COCO
| Backbone | size | SyncBN | AMP | Mem (GB) | box AP | Config | Download |
| :------: | :--: | :----: | :-: | :------: | :----: | :-------------------------------------------------------------------------------------------------------------------: | :------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: |
| YOLOv5-n | 640 | Yes | Yes | 1.5 | 28.0 | [config](https://github.com/open-mmlab/mmyolo/tree/master/configs/yolov5/yolov5_n-v61_syncbn_fast_8xb16-300e_coco.py) | [model](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_n-v61_syncbn_fast_8xb16-300e_coco/yolov5_n-v61_syncbn_fast_8xb16-300e_coco_20220919_090739-b804c1ad.pth) \| [log](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_n-v61_syncbn_fast_8xb16-300e_coco/yolov5_n-v61_syncbn_fast_8xb16-300e_coco_20220919_090739.log.json) |
| YOLOv5-s | 640 | Yes | Yes | 2.7 | 37.7 | [config](https://github.com/open-mmlab/mmyolo/tree/master/configs/yolov5/yolov5_s-v61_syncbn_fast_8xb16-300e_coco.py) | [model](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_s-v61_syncbn_fast_8xb16-300e_coco/yolov5_s-v61_syncbn_fast_8xb16-300e_coco_20220918_084700-86e02187.pth) \| [log](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_s-v61_syncbn_fast_8xb16-300e_coco/yolov5_s-v61_syncbn_fast_8xb16-300e_coco_20220918_084700.log.json) |
| YOLOv5-m | 640 | Yes | Yes | 5.0 | 45.3 | [config](https://github.com/open-mmlab/mmyolo/tree/master/configs/yolov5/yolov5_m-v61_syncbn_fast_8xb16-300e_coco.py) | [model](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_m-v61_syncbn_fast_8xb16-300e_coco/yolov5_m-v61_syncbn_fast_8xb16-300e_coco_20220917_204944-516a710f.pth) \| [log](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_m-v61_syncbn_fast_8xb16-300e_coco/yolov5_m-v61_syncbn_fast_8xb16-300e_coco_20220917_204944.log.json) |
| YOLOv5-l | 640 | Yes | Yes | 8.1 | 48.8 | [config](https://github.com/open-mmlab/mmyolo/tree/master/configs/yolov5/yolov5_l-v61_syncbn_fast_8xb16-300e_coco.py) | [model](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_l-v61_syncbn_fast_8xb16-300e_coco/yolov5_l-v61_syncbn_fast_8xb16-300e_coco_20220917_031007-096ef0eb.pth) \| [log](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_l-v61_syncbn_fast_8xb16-300e_coco/yolov5_l-v61_syncbn_fast_8xb16-300e_coco_20220917_031007.log.json) |
| Backbone | Arch | size | SyncBN | AMP | Mem (GB) | box AP | Config | Download |
| :------: | :--: | :--: | :----: | :-: | :------: | :----: | :----------------------------------------------------------------------------------------------------------------------: | :------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: |
| YOLOv5-n | P5 | 640 | Yes | Yes | 1.5 | 28.0 | [config](https://github.com/open-mmlab/mmyolo/tree/master/configs/yolov5/yolov5_n-v61_syncbn_fast_8xb16-300e_coco.py) | [model](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_n-v61_syncbn_fast_8xb16-300e_coco/yolov5_n-v61_syncbn_fast_8xb16-300e_coco_20220919_090739-b804c1ad.pth) \| [log](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_n-v61_syncbn_fast_8xb16-300e_coco/yolov5_n-v61_syncbn_fast_8xb16-300e_coco_20220919_090739.log.json) |
| YOLOv5-s | P5 | 640 | Yes | Yes | 2.7 | 37.7 | [config](https://github.com/open-mmlab/mmyolo/tree/master/configs/yolov5/yolov5_s-v61_syncbn_fast_8xb16-300e_coco.py) | [model](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_s-v61_syncbn_fast_8xb16-300e_coco/yolov5_s-v61_syncbn_fast_8xb16-300e_coco_20220918_084700-86e02187.pth) \| [log](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_s-v61_syncbn_fast_8xb16-300e_coco/yolov5_s-v61_syncbn_fast_8xb16-300e_coco_20220918_084700.log.json) |
| YOLOv5-m | P5 | 640 | Yes | Yes | 5.0 | 45.3 | [config](https://github.com/open-mmlab/mmyolo/tree/master/configs/yolov5/yolov5_m-v61_syncbn_fast_8xb16-300e_coco.py) | [model](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_m-v61_syncbn_fast_8xb16-300e_coco/yolov5_m-v61_syncbn_fast_8xb16-300e_coco_20220917_204944-516a710f.pth) \| [log](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_m-v61_syncbn_fast_8xb16-300e_coco/yolov5_m-v61_syncbn_fast_8xb16-300e_coco_20220917_204944.log.json) |
| YOLOv5-l | P5 | 640 | Yes | Yes | 8.1 | 48.8 | [config](https://github.com/open-mmlab/mmyolo/tree/master/configs/yolov5/yolov5_l-v61_syncbn_fast_8xb16-300e_coco.py) | [model](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_l-v61_syncbn_fast_8xb16-300e_coco/yolov5_l-v61_syncbn_fast_8xb16-300e_coco_20220917_031007-096ef0eb.pth) \| [log](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_l-v61_syncbn_fast_8xb16-300e_coco/yolov5_l-v61_syncbn_fast_8xb16-300e_coco_20220917_031007.log.json) |
| YOLOv5-n | P6 | 1280 | Yes | Yes | 5.8 | 35.9 | [config](https://github.com/open-mmlab/mmyolo/tree/master/configs/yolov5/yolov5_n-p6-v62_syncbn_fast_8xb16-300e_coco.py) | [model](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_n-p6-v62_syncbn_fast_8xb16-300e_coco/yolov5_n-p6-v62_syncbn_fast_8xb16-300e_coco_20221027_224705-d493c5f3.pth) \| [log](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_n-p6-v62_syncbn_fast_8xb16-300e_coco/yolov5_n-p6-v62_syncbn_fast_8xb16-300e_coco_20221027_224705.log.json) |
| YOLOv5-s | P6 | 1280 | Yes | Yes | 10.5 | 44.4 | [config](https://github.com/open-mmlab/mmyolo/tree/master/configs/yolov5/yolov5_s-p6-v62_syncbn_fast_8xb16-300e_coco.py) | [model](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_s-p6-v62_syncbn_fast_8xb16-300e_coco/yolov5_s-p6-v62_syncbn_fast_8xb16-300e_coco_20221027_215044-58865c19.pth) \| [log](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_s-p6-v62_syncbn_fast_8xb16-300e_coco/yolov5_s-p6-v62_syncbn_fast_8xb16-300e_coco_20221027_215044.log.json) |
| YOLOv5-m | P6 | 1280 | Yes | Yes | 19.1 | 51.3 | [config](https://github.com/open-mmlab/mmyolo/tree/master/configs/yolov5/yolov5_m-p6-v62_syncbn_fast_8xb16-300e_coco.py) | [model](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_m-p6-v62_syncbn_fast_8xb16-300e_coco/yolov5_m-p6-v62_syncbn_fast_8xb16-300e_coco_20221027_230453-49564d58.pth) \| [log](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_m-p6-v62_syncbn_fast_8xb16-300e_coco/yolov5_m-p6-v62_syncbn_fast_8xb16-300e_coco_20221027_230453.log.json) |
| YOLOv5-l | P6 | 1280 | Yes | Yes | 30.5 | 53.7 | [config](https://github.com/open-mmlab/mmyolo/tree/master/configs/yolov5/yolov5_l-p6-v62_syncbn_fast_8xb16-300e_coco.py) | [model](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_l-p6-v62_syncbn_fast_8xb16-300e_coco/yolov5_l-p6-v62_syncbn_fast_8xb16-300e_coco_20221027_234308-7a2ba6bf.pth) \| [log](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_l-p6-v62_syncbn_fast_8xb16-300e_coco/yolov5_l-p6-v62_syncbn_fast_8xb16-300e_coco_20221027_234308.log.json) |
**Note**:
In the official YOLOv5 code, the `random_perspective` data augmentation in COCO object detection task training uses mask annotation information, which leads to higher performance. Object detection should not use mask annotation, so only box annotation information is used in `MMYOLO`. We will use the mask annotation information in the instance segmentation task. See https://github.com/ultralytics/yolov5/issues/9917 for details.
1. `fast` means that `YOLOv5DetDataPreprocessor` and `yolov5_collate` are used for data preprocessing, which is faster for training, but less flexible for multitasking. Recommended to use fast version config if you only care about object detection.
2. `detect` means that the network input is fixed to `640x640` and the post-processing thresholds is modified.
3. `SyncBN` means use SyncBN, `AMP` indicates training with mixed precision.
4. We use 8x A100 for training, and the single-GPU batch size is 16. This is different from the official code.
5. The performance is unstable and may fluctuate by about 0.4 mAP.
5. The performance is unstable and may fluctuate by about 0.4 mAP and the highest performance weight in `COCO` training in `YOLOv5` may not be the last epoch.
6. `balloon` means that this is a demo configuration.
### VOC
@ -37,8 +42,6 @@ YOLOv5 is a family of object detection architectures and models pretrained on th
**Note**:
Please make sure `mmengine>=0.2.0`.
1. Training on VOC dataset need pretrained model which trained on COCO.
2. The performance is unstable and may fluctuate by about 0.4 mAP.
3. Official YOLOv5 use COCO metric, while training VOC dataset.

48
configs/yolov5/metafile.yml
View File

@ -80,6 +80,54 @@ Models:
Metrics:
box AP: 48.8
Weights: https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_l-v61_syncbn_fast_8xb16-300e_coco/yolov5_l-v61_syncbn_fast_8xb16-300e_coco_20220917_031007-096ef0eb.pth
- Name: yolov5_n-p6-v62_syncbn_fast_8xb16-300e_coco
In Collection: YOLOv5
Config: configs/yolov5/yolov5_n-p6-v62_syncbn_fast_8xb16-300e_coco.py
Metadata:
Training Memory (GB): 5.8
Epochs: 300
Results:
- Task: Object Detection
Dataset: COCO
Metrics:
box AP: 35.9
Weights: https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_n-p6-v62_syncbn_fast_8xb16-300e_coco/yolov5_n-p6-v62_syncbn_fast_8xb16-300e_coco_20221027_224705-d493c5f3.pth
- Name: yolov5_s-p6-v62_syncbn_fast_8xb16-300e_coco
In Collection: YOLOv5
Config: configs/yolov5/yolov5_s-p6-v62_syncbn_fast_8xb16-300e_coco.py
Metadata:
Training Memory (GB): 10.5
Epochs: 300
Results:
- Task: Object Detection
Dataset: COCO
Metrics:
box AP: 44.4
Weights: https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_s-p6-v62_syncbn_fast_8xb16-300e_coco/yolov5_s-p6-v62_syncbn_fast_8xb16-300e_coco_20221027_215044-58865c19.pth
- Name: yolov5_m-p6-v62_syncbn_fast_8xb16-300e_coco
In Collection: YOLOv5
Config: configs/yolov5/yolov5_m-p6-v62_syncbn_fast_8xb16-300e_coco.py
Metadata:
Training Memory (GB): 19.1
Epochs: 300
Results:
- Task: Object Detection
Dataset: COCO
Metrics:
box AP: 51.3
Weights: https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_m-p6-v62_syncbn_fast_8xb16-300e_coco/yolov5_m-p6-v62_syncbn_fast_8xb16-300e_coco_20221027_230453-49564d58.pth
- Name: yolov5_l-p6-v62_syncbn_fast_8xb16-300e_coco
In Collection: YOLOv5
Config: configs/yolov5/yolov5_l-p6-v62_syncbn_fast_8xb16-300e_coco.py
Metadata:
Training Memory (GB): 30.5
Epochs: 300
Results:
- Task: Object Detection
Dataset: COCO
Metrics:
box AP: 53.7
Weights: https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_l-p6-v62_syncbn_fast_8xb16-300e_coco/yolov5_l-p6-v62_syncbn_fast_8xb16-300e_coco_20221027_234308-7a2ba6bf.pth
- Name: yolov5_n-v61_fast_1xb64-50e_voc
In Collection: YOLOv5_VOC
Config: configs/yolov5/voc/yolov5_n-v61_fast_1xb64-50e_voc.py

15
configs/yolov5/yolov5_l-p6-v62_syncbn_fast_8xb16-300e_coco.py 100644
View File

@ -0,0 +1,15 @@
_base_ = './yolov5_m-p6-v62_syncbn_fast_8xb16-300e_coco.py'
deepen_factor = 1.0
widen_factor = 1.0
model = dict(
backbone=dict(
deepen_factor=deepen_factor,
widen_factor=widen_factor,
),
neck=dict(
deepen_factor=deepen_factor,
widen_factor=widen_factor,
),
bbox_head=dict(head_module=dict(widen_factor=widen_factor)))

74
configs/yolov5/yolov5_m-p6-v62_syncbn_fast_8xb16-300e_coco.py 100644
View File

@ -0,0 +1,74 @@
_base_ = './yolov5_s-p6-v62_syncbn_fast_8xb16-300e_coco.py'
deepen_factor = 0.67
widen_factor = 0.75
lr_factor = 0.1 # lrf=0.1
affine_scale = 0.9
num_classes = _base_.num_classes
num_det_layers = _base_.num_det_layers
img_scale = _base_.img_scale
model = dict(
backbone=dict(
deepen_factor=deepen_factor,
widen_factor=widen_factor,
),
neck=dict(
deepen_factor=deepen_factor,
widen_factor=widen_factor,
),
bbox_head=dict(
head_module=dict(widen_factor=widen_factor),
loss_cls=dict(loss_weight=0.3 *
(num_classes / 80 * 3 / num_det_layers)),
loss_obj=dict(loss_weight=0.7 *
((img_scale[0] / 640)**2 * 3 / num_det_layers))))
pre_transform = _base_.pre_transform
albu_train_transforms = _base_.albu_train_transforms
mosaic_affine_pipeline = [
dict(
type='Mosaic',
img_scale=img_scale,
pad_val=114.0,
pre_transform=pre_transform),
dict(
type='YOLOv5RandomAffine',
max_rotate_degree=0.0,
max_shear_degree=0.0,
scaling_ratio_range=(1 - affine_scale, 1 + affine_scale),
border=(-img_scale[0] // 2, -img_scale[1] // 2),
border_val=(114, 114, 114))
]
# enable mixup
train_pipeline = [
*pre_transform, *mosaic_affine_pipeline,
dict(
type='YOLOv5MixUp',
prob=0.1,
pre_transform=[*pre_transform, *mosaic_affine_pipeline]),
dict(
type='mmdet.Albu',
transforms=albu_train_transforms,
bbox_params=dict(
type='BboxParams',
format='pascal_voc',
label_fields=['gt_bboxes_labels', 'gt_ignore_flags']),
keymap={
'img': 'image',
'gt_bboxes': 'bboxes'
}),
dict(type='YOLOv5HSVRandomAug'),
dict(type='mmdet.RandomFlip', prob=0.5),
dict(
type='mmdet.PackDetInputs',
meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape', 'flip',
'flip_direction'))
]
train_dataloader = dict(dataset=dict(pipeline=train_pipeline))
default_hooks = dict(param_scheduler=dict(lr_factor=lr_factor))

11
configs/yolov5/yolov5_m-v61_syncbn_fast_8xb16-300e_coco.py
View File

@ -5,6 +5,10 @@ widen_factor = 0.75
lr_factor = 0.1 # lrf=0.1
affine_scale = 0.9
num_classes = _base_.num_classes
num_det_layers = _base_.num_det_layers
img_scale = _base_.img_scale
model = dict(
backbone=dict(
deepen_factor=deepen_factor,
@ -16,12 +20,13 @@ model = dict(
),
bbox_head=dict(
head_module=dict(widen_factor=widen_factor),
loss_cls=dict(loss_weight=0.3),
loss_obj=dict(loss_weight=0.7)))
loss_cls=dict(loss_weight=0.3 *
(num_classes / 80 * 3 / num_det_layers)),
loss_obj=dict(loss_weight=0.7 *
((img_scale[0] / 640)**2 * 3 / num_det_layers))))
pre_transform = _base_.pre_transform
albu_train_transforms = _base_.albu_train_transforms
img_scale = _base_.img_scale
mosaic_affine_pipeline = [
dict(

15
configs/yolov5/yolov5_n-p6-v62_syncbn_fast_8xb16-300e_coco.py 100644
View File

@ -0,0 +1,15 @@
_base_ = 'yolov5_s-p6-v62_syncbn_fast_8xb16-300e_coco.py'
deepen_factor = 0.33
widen_factor = 0.25
model = dict(
backbone=dict(
deepen_factor=deepen_factor,
widen_factor=widen_factor,
),
neck=dict(
deepen_factor=deepen_factor,
widen_factor=widen_factor,
),
bbox_head=dict(head_module=dict(widen_factor=widen_factor)))

89
configs/yolov5/yolov5_s-p6-v62_syncbn_fast_8xb16-300e_coco.py 100644
View File

@ -0,0 +1,89 @@
_base_ = 'yolov5_s-v61_syncbn_fast_8xb16-300e_coco.py'
img_scale = (1280, 1280) # height, width
num_classes = 80
# only on Val
batch_shapes_cfg = dict(img_size=img_scale[0], size_divisor=64)
anchors = [
[(19, 27), (44, 40), (38, 94)], # P3/8
[(96, 68), (86, 152), (180, 137)], # P4/16
[(140, 301), (303, 264), (238, 542)], # P5/32
[(436, 615), (739, 380), (925, 792)] # P6/64
]
strides = [8, 16, 32, 64]
num_det_layers = 4
model = dict(
backbone=dict(arch='P6', out_indices=(2, 3, 4, 5)),
neck=dict(
in_channels=[256, 512, 768, 1024], out_channels=[256, 512, 768, 1024]),
bbox_head=dict(
head_module=dict(
in_channels=[256, 512, 768, 1024], featmap_strides=strides),
prior_generator=dict(base_sizes=anchors, strides=strides),
# scaled based on number of detection layers
loss_cls=dict(loss_weight=0.5 *
(num_classes / 80 * 3 / num_det_layers)),
loss_bbox=dict(loss_weight=0.05 * (3 / num_det_layers)),
loss_obj=dict(loss_weight=1.0 *
((img_scale[0] / 640)**2 * 3 / num_det_layers)),
obj_level_weights=[4.0, 1.0, 0.25, 0.06]))
pre_transform = _base_.pre_transform
albu_train_transforms = _base_.albu_train_transforms
train_pipeline = [
*pre_transform,
dict(
type='Mosaic',
img_scale=img_scale,
pad_val=114.0,
pre_transform=pre_transform),
dict(
type='YOLOv5RandomAffine',
max_rotate_degree=0.0,
max_shear_degree=0.0,
scaling_ratio_range=(0.5, 1.5),
border=(-img_scale[0] // 2, -img_scale[1] // 2),
border_val=(114, 114, 114)),
dict(
type='mmdet.Albu',
transforms=albu_train_transforms,
bbox_params=dict(
type='BboxParams',
format='pascal_voc',
label_fields=['gt_bboxes_labels', 'gt_ignore_flags']),
keymap={
'img': 'image',
'gt_bboxes': 'bboxes'
}),
dict(type='YOLOv5HSVRandomAug'),
dict(type='mmdet.RandomFlip', prob=0.5),
dict(
type='mmdet.PackDetInputs',
meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape', 'flip',
'flip_direction'))
]
train_dataloader = dict(dataset=dict(pipeline=train_pipeline))
test_pipeline = [
dict(type='LoadImageFromFile', file_client_args=_base_.file_client_args),
dict(type='YOLOv5KeepRatioResize', scale=img_scale),
dict(
type='LetterResize',
scale=img_scale,
allow_scale_up=False,
pad_val=dict(img=114)),
dict(type='LoadAnnotations', with_bbox=True, _scope_='mmdet'),
dict(
type='mmdet.PackDetInputs',
meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape',
'scale_factor', 'pad_param'))
]
val_dataloader = dict(
dataset=dict(pipeline=test_pipeline, batch_shapes_cfg=batch_shapes_cfg))
test_dataloader = val_dataloader

31
configs/yolov5/yolov5_s-v61_syncbn_8xb16-300e_coco.py
View File

@ -5,6 +5,7 @@ data_root = 'data/coco/'
dataset_type = 'YOLOv5CocoDataset'
# parameters that often need to be modified
num_classes = 80
img_scale = (640, 640) # height, width
deepen_factor = 0.33
widen_factor = 0.5
@ -26,9 +27,13 @@ batch_shapes_cfg = dict(
size_divisor=32,
extra_pad_ratio=0.5)
anchors = [[(10, 13), (16, 30), (33, 23)], [(30, 61), (62, 45), (59, 119)],
[(116, 90), (156, 198), (373, 326)]]
anchors = [
[(10, 13), (16, 30), (33, 23)], # P3/8
[(30, 61), (62, 45), (59, 119)], # P4/16
[(116, 90), (156, 198), (373, 326)] # P5/32
]
strides = [8, 16, 32]
num_det_layers = 3
# single-scale training is recommended to
# be turned on, which can speed up training.
@ -60,7 +65,7 @@ model = dict(
type='YOLOv5Head',
head_module=dict(
type='YOLOv5HeadModule',
num_classes=80,
num_classes=num_classes,
in_channels=[256, 512, 1024],
widen_factor=widen_factor,
featmap_strides=strides,
@ -69,24 +74,25 @@ model = dict(
type='mmdet.YOLOAnchorGenerator',
base_sizes=anchors,
strides=strides),
# scaled based on number of detection layers
loss_cls=dict(
type='mmdet.CrossEntropyLoss',
use_sigmoid=True,
reduction='mean',
loss_weight=0.5),
loss_weight=0.5 * (num_classes / 80 * 3 / num_det_layers)),
loss_bbox=dict(
type='IoULoss',
iou_mode='ciou',
bbox_format='xywh',
eps=1e-7,
reduction='mean',
loss_weight=0.05,
loss_weight=0.05 * (3 / num_det_layers),
return_iou=True),
loss_obj=dict(
type='mmdet.CrossEntropyLoss',
use_sigmoid=True,
reduction='mean',
loss_weight=1.0),
loss_weight=1.0 * ((img_scale[0] / 640)**2 * 3 / num_det_layers)),
prior_match_thr=4.,
obj_level_weights=[4., 1., 0.4]),
test_cfg=dict(
@ -104,9 +110,7 @@ albu_train_transforms = [
]
pre_transform = [
dict(
type='LoadImageFromFile',
file_client_args={{_base_.file_client_args}}),
dict(type='LoadImageFromFile', file_client_args=_base_.file_client_args),
dict(type='LoadAnnotations', with_bbox=True)
]
@ -158,9 +162,7 @@ train_dataloader = dict(
pipeline=train_pipeline))
test_pipeline = [
dict(
type='LoadImageFromFile',
file_client_args={{_base_.file_client_args}}),
dict(type='LoadImageFromFile', file_client_args=_base_.file_client_args),
dict(type='YOLOv5KeepRatioResize', scale=img_scale),
dict(
type='LetterResize',
@ -211,7 +213,9 @@ default_hooks = dict(
lr_factor=0.01,
max_epochs=max_epochs),
checkpoint=dict(
type='CheckpointHook', interval=save_epoch_intervals,
type='CheckpointHook',
interval=save_epoch_intervals,
save_best='auto',
max_keep_ckpts=3))
custom_hooks = [
@ -220,6 +224,7 @@ custom_hooks = [
ema_type='ExpMomentumEMA',
momentum=0.0001,
update_buffers=True,
strict_load=False,
priority=49)
]

14
configs/yolov5/yolov5_x-p6-v62_syncbn_fast_8xb16-300e_coco.py 100644
View File

@ -0,0 +1,14 @@
_base_ = './yolov5_m-p6-v62_syncbn_fast_8xb16-300e_coco.py'
deepen_factor = 1.33
widen_factor = 1.25
model = dict(
backbone=dict(
deepen_factor=deepen_factor,
widen_factor=widen_factor,
),
neck=dict(
deepen_factor=deepen_factor,
widen_factor=widen_factor,
),
bbox_head=dict(head_module=dict(widen_factor=widen_factor)))

1
configs/yolox/yolox_s_8xb8-300e_coco.py
View File

@ -264,6 +264,7 @@ custom_hooks = [
ema_type='ExpMomentumEMA',
momentum=0.0001,
update_buffers=True,
strict_load=False,
priority=49)
]

2
docker/Dockerfile
View File

@ -26,7 +26,7 @@ RUN apt-get update \
# Install MMEngine , MMCV and MMDet
RUN pip install --no-cache-dir openmim && \
mim install --no-cache-dir "mmengine>=0.2.0" "mmcv>=2.0.0rc1,<2.1.0" "mmdet>=3.0.0rc1,<3.1.0"
mim install --no-cache-dir "mmengine>=0.2.0" "mmcv>=2.0.0rc1,<2.1.0" "mmdet>=3.0.0rc2,<3.1.0"
# Install MMYOLO
RUN git clone https://github.com/open-mmlab/mmyolo.git /mmyolo && \

8
docs/en/get_started.md
View File

@ -6,10 +6,10 @@ Compatible MMEngine, MMCV and MMDetection versions are shown as below. Please in
| MMYOLO version | MMDetection version | MMEngine version | MMCV version |
| :------------: | :----------------------: | :----------------------: | :---------------------: |
| main | mmdet>=3.0.0rc1, \<3.1.0 | mmengine>=0.2.0, \<1.0.0 | mmcv>=2.0.0rc0, \<2.1.0 |
| 0.1.2 | mmdet>=3.0.0rc1, \<3.1.0 | mmengine>=0.2.0, \<1.0.0 | mmcv>=2.0.0rc0, \<2.1.0 |
| 0.1.1 | mmdet>=3.0.0rc1, \<3.1.0 | mmengine>=0.1.0, \<0.2.0 | mmcv>=2.0.0rc0, \<2.1.0 |
| 0.1.0 | mmdet>=3.0.0rc0, \<3.1.0 | mmengine>=0.1.0, \<0.2.0 | mmcv>=2.0.0rc0, \<2.1.0 |
| main | mmdet>=3.0.0rc2, \<3.1.0 | mmengine>=0.2.0, \<1.0.0 | mmcv>=2.0.0rc0, \<2.1.0 |
| 0.1.2 | mmdet>=3.0.0rc2, \<3.1.0 | mmengine>=0.2.0, \<1.0.0 | mmcv>=2.0.0rc0, \<2.1.0 |
| 0.1.1 | mmdet==3.0.0rc1 | mmengine>=0.1.0, \<0.2.0 | mmcv>=2.0.0rc0, \<2.1.0 |
| 0.1.0 | mmdet==3.0.0rc0 | mmengine>=0.1.0, \<0.2.0 | mmcv>=2.0.0rc0, \<2.1.0 |
In this section, we demonstrate how to prepare an environment with PyTorch.

3
docs/en/user_guides/yolov5_tutorial.md
View File

@ -12,8 +12,7 @@ conda install pytorch torchvision -c pytorch
pip install -U openmim
mim install "mmengine>=0.2.0"
mim install "mmcv>=2.0.0rc1,<2.1.0"
mim install "mmdet>=3.0.0rc1,<3.1.0"
# for albumentations
mim install "mmdet>=3.0.0rc2,<3.1.0"
git clone https://github.com/open-mmlab/mmyolo.git
cd mmyolo
# Install albumentations

8
docs/zh_cn/get_started.md
View File

@ -6,10 +6,10 @@
| MMYOLO version | MMDetection version | MMEngine version | MMCV version |
| :------------: | :----------------------: | :----------------------: | :---------------------: |
| main | mmdet>=3.0.0rc1, \<3.1.0 | mmengine>=0.2.0, \<1.0.0 | mmcv>=2.0.0rc0, \<2.1.0 |
| 0.1.2 | mmdet>=3.0.0rc1, \<3.1.0 | mmengine>=0.2.0, \<1.0.0 | mmcv>=2.0.0rc0, \<2.1.0 |
| 0.1.1 | mmdet>=3.0.0rc1, \<3.1.0 | mmengine>=0.1.0, \<0.2.0 | mmcv>=2.0.0rc0, \<2.1.0 |
| 0.1.0 | mmdet>=3.0.0rc0, \<3.1.0 | mmengine>=0.1.0, \<0.2.0 | mmcv>=2.0.0rc0, \<2.1.0 |
| main | mmdet>=3.0.0rc2, \<3.1.0 | mmengine>=0.2.0, \<1.0.0 | mmcv>=2.0.0rc0, \<2.1.0 |
| 0.1.2 | mmdet>=3.0.0rc2, \<3.1.0 | mmengine>=0.2.0, \<1.0.0 | mmcv>=2.0.0rc0, \<2.1.0 |
| 0.1.1 | mmdet==3.0.0rc1 | mmengine>=0.1.0, \<0.2.0 | mmcv>=2.0.0rc0, \<2.1.0 |
| 0.1.0 | mmdet==3.0.0rc0 | mmengine>=0.1.0, \<0.2.0 | mmcv>=2.0.0rc0, \<2.1.0 |
本节中，我们将演示如何用 PyTorch 准备一个环境。

3
docs/zh_cn/user_guides/yolov5_tutorial.md
View File

@ -12,8 +12,7 @@ conda install pytorch torchvision -c pytorch
pip install -U openmim
mim install "mmengine>=0.2.0"
mim install "mmcv>=2.0.0rc1,<2.1.0"
mim install "mmdet>=3.0.0rc1,<3.1.0"
# for albumentations
mim install "mmdet>=3.0.0rc2,<3.1.0"
git clone https://github.com/open-mmlab/mmyolo.git
cd mmyolo
# Install albumentations

2
mmyolo/__init__.py
View File

@ -14,7 +14,7 @@ mmengine_minimum_version = '0.2.0'
mmengine_maximum_version = '1.0.0'
mmengine_version = digit_version(mmengine.__version__)
mmdet_minimum_version = '3.0.0rc1'
mmdet_minimum_version = '3.0.0rc2'
mmdet_maximum_version = '3.1.0'
mmdet_version = digit_version(mmdet.__version__)

5
mmyolo/models/backbones/csp_darknet.py
View File

@ -62,7 +62,10 @@ class YOLOv5CSPDarknet(BaseBackbone):
# in_channels, out_channels, num_blocks, add_identity, use_spp
arch_settings = {
'P5': [[64, 128, 3, True, False], [128, 256, 6, True, False],
[256, 512, 9, True, False], [512, 1024, 3, True, True]]
[256, 512, 9, True, False], [512, 1024, 3, True, True]],
'P6': [[64, 128, 3, True, False], [128, 256, 6, True, False],
[256, 512, 9, True, False], [512, 768, 3, True, False],
[768, 1024, 3, True, True]]
}
def __init__(self,

5
mmyolo/models/necks/yolov5_pafpn.py
View File

@ -72,7 +72,7 @@ class YOLOv5PAFPN(BaseYOLONeck):
Returns:
nn.Module: The reduce layer.
"""
if idx == 2:
if idx == len(self.in_channels) - 1:
layer = ConvModule(
make_divisible(self.in_channels[idx], self.widen_factor),
make_divisible(self.in_channels[idx - 1], self.widen_factor),
@ -97,6 +97,7 @@ class YOLOv5PAFPN(BaseYOLONeck):
Returns:
nn.Module: The top down layer.
"""
if idx == 1:
return CSPLayer(
make_divisible(self.in_channels[idx - 1] * 2,
@ -106,7 +107,7 @@ class YOLOv5PAFPN(BaseYOLONeck):
add_identity=False,
norm_cfg=self.norm_cfg,
act_cfg=self.act_cfg)
elif idx == 2:
else:
return nn.Sequential(
CSPLayer(
make_divisible(self.in_channels[idx - 1] * 2,

2
requirements/mminstall.txt
View File

@ -1,3 +1,3 @@
mmcv>=2.0.0rc1,<2.1.0
mmdet>=3.0.0rc1
mmdet>=3.0.0rc2
mmengine>=0.2.0

2
requirements/readthedocs.txt
View File

@ -1,5 +1,5 @@
mmcv>=2.0.0rc1,<2.1.0
mmdet>=3.0.0rc1
mmdet>=3.0.0rc2
mmengine>=0.2.0
torch
torchvision

57
tools/model_converters/yolov5_to_mmyolo.py
View File

@ -4,7 +4,7 @@ from collections import OrderedDict
import torch
convert_dict = {
convert_dict_p5 = {
'model.0': 'backbone.stem',
'model.1': 'backbone.stage1.0',
'model.2': 'backbone.stage1.1',
@ -27,17 +27,62 @@ convert_dict = {
'model.24.m': 'bbox_head.head_module.convs_pred',
}
convert_dict_p6 = {
'model.0': 'backbone.stem',
'model.1': 'backbone.stage1.0',
'model.2': 'backbone.stage1.1',
'model.3': 'backbone.stage2.0',
'model.4': 'backbone.stage2.1',
'model.5': 'backbone.stage3.0',
'model.6': 'backbone.stage3.1',
'model.7': 'backbone.stage4.0',
'model.8': 'backbone.stage4.1',
'model.9': 'backbone.stage5.0',
'model.10': 'backbone.stage5.1',
'model.11.cv1': 'backbone.stage5.2.conv1',
'model.11.cv2': 'backbone.stage5.2.conv2',
'model.12': 'neck.reduce_layers.3',
'model.15': 'neck.top_down_layers.0.0',
'model.16': 'neck.top_down_layers.0.1',
'model.19': 'neck.top_down_layers.1.0',
'model.20': 'neck.top_down_layers.1.1',
'model.23': 'neck.top_down_layers.2',
'model.24': 'neck.downsample_layers.0',
'model.26': 'neck.bottom_up_layers.0',
'model.27': 'neck.downsample_layers.1',
'model.29': 'neck.bottom_up_layers.1',
'model.30': 'neck.downsample_layers.2',
'model.32': 'neck.bottom_up_layers.2',
'model.33.m': 'bbox_head.head_module.convs_pred',
}
def convert(src, dst):
"""Convert keys in detectron pretrained YOLOV5 models to mmyolo style."""
yolov5_model = torch.load(src)['model']
blobs = yolov5_model.state_dict()
"""Convert keys in pretrained YOLOv5 models to mmyolo style."""
if src.endswith('6.pt'):
convert_dict = convert_dict_p6
is_p6_model = True
print('Converting P6 model')
else:
convert_dict = convert_dict_p5
is_p6_model = False
print('Converting P5 model')
try:
yolov5_model = torch.load(src)['model']
blobs = yolov5_model.state_dict()
except ModuleNotFoundError:
raise RuntimeError(
'This script must be placed under the ultralytics/yolov5 repo,'
' because loading the official pretrained model need'
' `model.py` to build model.')
state_dict = OrderedDict()
for key, weight in blobs.items():
num, module = key.split('.')[1:3]
if num == '9' or num == '24':
if (is_p6_model and
(num == '11' or num == '33')) or (not is_p6_model and
(num == '9' or num == '24')):
if module == 'anchors':
continue
prefix = f'model.{num}.{module}'
@ -68,7 +113,7 @@ def main():
parser = argparse.ArgumentParser(description='Convert model keys')
parser.add_argument(
'--src', default='yolov5s.pt', help='src yolov5 model path')
parser.add_argument('--dst', default='mmyolov5.pt', help='save path')
parser.add_argument('--dst', default='mmyolov5s.pt', help='save path')
args = parser.parse_args()
convert(args.src, args.dst)