[Enhancement] RepVGG for YOLOX-PAI

[okotaku]

Motivation

Add MTSPPF block and add_ppf param for YOLOX-PAI.
YOLOX-PAI: An Improved YOLOX, Stronger and Faster than YOLOv6
official repo

Modification

1. add MTSPPF module and support it in RepVGG backbone;
2. support 'stem_channels' in arch and add a 'yolox-pai-small' arch cfg
3. Reduce the number of channels to speed up unit testing

Related PR

open-mmlab/mmdetection#8778

Checklist

Before PR:

• Pre-commit or other linting tools are used to fix the potential lint issues.
• Bug fixes are fully covered by unit tests, the case that causes the bug should be added in the unit tests.
• The modification is covered by complete unit tests. If not, please add more unit test to ensure the correctness.
• The documentation has been modified accordingly, like docstring or example tutorials.

After PR:

• If the modification has potential influence on downstream or other related projects, this PR should be tested with those projects, like MMDet or MMSeg.
• CLA has been signed and all committers have signed the CLA in this PR.

[codecov]

Codecov Report

Base: 86.09% // Head: 86.09% // Decreases project coverage by -0.00% ⚠️

Coverage data is based on head (e4881a6) compared to base (517bd3d).
Patch coverage: 100.00% of modified lines in pull request are covered.

Additional details and impacted files

@@            Coverage Diff             @@
##              dev    #1025      +/-   ##
==========================================
- Coverage   86.09%   86.09%   -0.01%     
==========================================
  Files         140      142       +2     
  Lines        9676     9896     +220     
  Branches     1677     1713      +36     
==========================================
+ Hits         8331     8520     +189     
- Misses       1092     1114      +22     
- Partials      253      262       +9     

Flag	Coverage Δ
unittests	86.02% <100.00%> (-0.03%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
mmcls/models/backbones/repvgg.py	96.15% <100.00%> (-0.69%)	⬇️
mmcls/apis/train.py	15.38% <0.00%> (ø)
mmcls/models/backbones/van.py	96.83% <0.00%> (ø)
mmcls/models/utils/__init__.py	100.00% <0.00%> (ø)
mmcls/models/utils/attention.py	96.91% <0.00%> (ø)
mmcls/models/backbones/__init__.py	100.00% <0.00%> (ø)
mmcls/models/backbones/hornet.py	83.07% <0.00%> (ø)
mmcls/models/utils/layer_scale.py	86.66% <0.00%> (ø)
mmcls/models/backbones/t2t_vit.py	95.18% <0.00%> (+0.05%)	⬆️
... and 3 more

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report at Codecov.
📢 Do you have feedback about the report comment? Let us know in this issue.

[Ezra-Yu]

>>> from mmcls.models.backbones import RepVGG
>>> model = RepVGG('yolox-pai-small', out_indices=(1, 2, 3, ), add_ppf=True )

Is this model what you want? @okotaku

[okotaku]

@Ezra-Yu Yes it is!
I fixed mmdet PR. https://github.com/RistoranteRist/mmdetection/blob/yoloxpai/configs/yoloxpai/yolox_pai_s_8x8_300e_coco.py#L8

[Ezra-Yu]

LGTM.

[mzr1996]

RepVGG architecture.