[Fix] Fix SpatialReductionAttention in PVT.

[RangiLyu]

Thanks for your contribution and we appreciate it a lot. The following instructions would make your pull request more healthy and more easily get feedback. If you do not understand some items, don't worry, just make the pull request and seek help from maintainers.

Motivation

The latest MMCV(v1.3.17) include a commit open-mmlab/mmcv#1418, which break the compatibility of SpatialReductionAttention in PVT.

Modification

Change the forward of SpatialReductionAttention and add a legacy_forward.

Handle the mmcv version by this:

  # handle the BC-breaking from https://github.com/open-mmlab/mmcv/pull/1418 # noqa
  from mmdet import mmcv_version, digit_version
  if mmcv_version < digit_version('1.3.17'):
      self.forward = self.legacy_forward

BC-breaking (Optional)

None

Checklist

1. Pre-commit or other linting tools are used to fix the potential lint issues.
2. The modification is covered by complete unit tests. If not, please add more unit test to ensure the correctness.
3. If the modification has potential influence on downstream projects, this PR should be tested with downstream projects, like MMDet or MMCls.
4. The documentation has been modified accordingly, like docstring or example tutorials.

[codecov]

Codecov Report

Merging #6488 (5327663) into master (9874180) will increase coverage by 0.25%.
The diff coverage is 43.87%.

@@            Coverage Diff             @@
##           master    #6488      +/-   ##
==========================================
+ Coverage   61.80%   62.05%   +0.25%     
==========================================
  Files         315      319       +4     
  Lines       25174    25335     +161     
  Branches     4187     4194       +7     
==========================================
+ Hits        15558    15721     +163     
+ Misses       8805     8783      -22     
- Partials      811      831      +20     

Flag	Coverage Δ
unittests	62.02% <43.87%> (+0.24%)	⬆️

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

Impacted Files	Coverage Δ
mmdet/apis/train.py	15.90% <0.00%> (-0.31%)	⬇️
mmdet/models/roi_heads/mask_heads/fcn_mask_head.py	64.73% <0.00%> (-0.38%)	⬇️
mmdet/datasets/builder.py	55.81% <7.69%> (-5.73%)	⬇️
mmdet/models/roi_heads/sparse_roi_head.py	70.06% <18.75%> (-24.17%)	⬇️
mmdet/datasets/samplers/infinite_sampler.py	23.75% <23.75%> (ø)
mmdet/models/backbones/pvt.py	77.99% <33.33%> (-5.61%)	⬇️
mmdet/models/dense_heads/anchor_head.py	86.98% <50.00%> (+0.25%)	⬆️
mmdet/models/detectors/queryinst.py	83.33% <83.33%> (ø)
...t/models/roi_heads/mask_heads/dynamic_mask_head.py	83.33% <83.33%> (ø)
mmdet/datasets/pipelines/transforms.py	77.09% <100.00%> (ø)
... and 49 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 e065958...5327663. Read the comment docs.