Make YOLOv3 neck more flexible

[hokmund]

Motivation

At the moment, YOLOV3Neck only allows to create neck architectures that have out_channels[0] == in_channels[1] and out_channels[1] == i_channels[2]. This is due to this line of code:

        for i in range(1, self.num_scales):
            in_c, out_c = self.in_channels[i], self.out_channels[i]
            self.add_module(f'conv{i}', ConvModule(in_c, out_c, 1, **cfg)) <<<---
            # in_c + out_c : High-lvl feats will be cat with low-lvl feats
            self.add_module(f'detect{i+1}',
                            DetectionBlock(in_c + out_c, out_c, **cfg))

It can be easily fixed to allow more flexible architectures without any backward compatibility issues.

It is convenient when you assemble YOLO-like architecture using some pretrained backbone, e.g. mobilenetv2_100 that has [320, 96, 32] channels in output feature maps. Current implementation enforces you to use very tight neck with [96, 32, X] channels. Less strict implementation will allow you to create a wider neck and increase your model's capacity.

Modification

Change input channels of conv{i} layers in order to accept arbitrary number of output channels from the previous DetectionBlock.

BC-breaking (Optional)

No breaking changes.

[codecov]

Codecov Report

Merging #5218 (ae26378) into master (ade08ba) will decrease coverage by 1.13%.
The diff coverage is 39.80%.

❗ Current head ae26378 differs from pull request most recent head b962b6e. Consider uploading reports for the commit b962b6e to get more accurate results

@@            Coverage Diff             @@
##           master    #5218      +/-   ##
==========================================
- Coverage   65.39%   64.25%   -1.14%     
==========================================
  Files         276      275       -1     
  Lines       21229    21094     -135     
  Branches     3561     3532      -29     
==========================================
- Hits        13883    13555     -328     
- Misses       6572     6773     +201     
+ Partials      774      766       -8     

Flag	Coverage Δ
unittests	64.25% <39.80%> (-1.11%)	⬇️

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

Impacted Files	Coverage Δ
mmdet/core/export/model_wrappers.py	0.00% <0.00%> (ø)
mmdet/core/export/pytorch2onnx.py	15.68% <ø> (ø)
mmdet/datasets/coco.py	45.84% <ø> (ø)
mmdet/models/dense_heads/anchor_head.py	83.62% <ø> (-3.10%)	⬇️
mmdet/models/dense_heads/cascade_rpn_head.py	22.71% <0.00%> (ø)
mmdet/models/detectors/detr.py	100.00% <ø> (ø)
mmdet/models/roi_heads/point_rend_roi_head.py	19.62% <ø> (ø)
...heads/roi_extractors/single_level_roi_extractor.py	65.38% <0.00%> (-4.01%)	⬇️
mmdet/models/roi_heads/standard_roi_head.py	50.31% <7.31%> (-12.09%)	⬇️
mmdet/models/dense_heads/yolo_head.py	27.11% <18.18%> (-24.05%)	⬇️
... and 89 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 7d49b7b...b962b6e. Read the comment docs.

[RangiLyu]

Thanks for your contribution!

[ZwwWayne]

Can be merged after fixing the docstring format issues.

[OpenMMLab-Coodinator]

Hi @hokmund ！First of all, we want to express our gratitude for your significant PR in this project. Your contribution is highly appreciated, and we are grateful for your efforts in helping improve this open-source project during your personal time. We believe that many developers will benefit from your PR.

We would also like to invite you to join our Special Interest Group (SIG) private channel on Discord, where you can share your experiences, ideas, and build connections with like-minded peers. To join the SIG channel, simply message moderator— OpenMMLab on Discord or briefly share your open-source contributions in the #introductions channel and we will assist you. Look forward to seeing you there! Join us ：https://discord.gg/UjgXkPWNqA

If you have WeChat account，welcome to join our community on WeChat. You can add our assistant ：openmmlabwx. Please add "mmsig + Github ID" as a remark when adding friends：）
Thank you again for your contribution❤