Add PVTv2 backbone to RetinaNet

[dnth]

This PR adds Pyramid Vision Transformer v2 backbone to the RetinaNet model.

Configs and weights from mmdet repo https://github.com/open-mmlab/mmdetection/tree/master/configs/pvt

[codecov[bot]]

Codecov Report

Merging #1043 (bcd3180) into master (bfbed41) will decrease coverage by 0.01%. The diff coverage is 95.45%.

@@            Coverage Diff             @@
##           master    #1043      +/-   ##
==========================================
- Coverage   85.96%   85.95%   -0.02%     
==========================================
  Files         272      282      +10     
  Lines        5794     5939     +145     
==========================================
+ Hits         4981     5105     +124     
- Misses        813      834      +21     

Flag	Coverage Δ
unittests	85.95% <95.45%> (-0.02%)	:arrow_down:

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

Impacted Files	Coverage Δ
icevision/models/mmdet/utils.py	57.35% <50.00%> (-4.32%)	:arrow_down:
icevision/models/mmdet/download_configs.py	80.95% <100.00%> (ø)
...odels/mmdet/models/retinanet/backbones/__init__.py	100.00% <100.00%> (ø)
...mdet/models/retinanet/backbones/backbone_config.py	100.00% <100.00%> (ø)
...ion/models/mmdet/models/retinanet/backbones/pvt.py	100.00% <100.00%> (ø)
...els/mmdet/models/retinanet/backbones/resnet_fpn.py	100.00% <100.00%> (ø)
...on/models/mmdet/models/retinanet/backbones/swin.py	100.00% <100.00%> (ø)
...on/models/mmdet/models/vfnet/backbones/__init__.py	100.00% <100.00%> (ø)
...ls/mmdet/models/vfnet/backbones/backbone_config.py	100.00% <100.00%> (ø)
.../models/mmdet/models/vfnet/backbones/resnet_fpn.py	100.00% <100.00%> (ø)
... and 12 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 bfbed41...bcd3180. Read the comment docs.

[ai-fast-track]

Hey @dnth,

I just wanted to confirm this PR is aligned with the structure that we designed during our call.

I have the impression the answer is Yes. In that case, I can merged it!

[dnth]

Hey @dnth,

I just wanted to confirm this PR is aligned with the structure that we designed during our call.

I have the impression the answer is Yes. In that case, I can merged it!

Now it is!

[dnth]

COCO Metric for RetinaNet with PVTv2 B3 backbone

[dnth]

To use PVT backbone we must update to mmdet==2.20.0. However this causes a problem with VFNet models.