[Fix] Fix CI bug due to mmcv index

[kennymckormick]

No description provided.

[innerlee]

-q seems unrelated.

[kennymckormick]

-q seems unrelated.

just to check log, seems if -q added, nothing will be printed

[innerlee]

Is this caused by mmdet? The CI was successful before

[kennymckormick]

Is this caused by mmdet? The CI was successful before

Seems the mmcv with wrong version is installed in the cuda CI, see 2 screenshots below:

Try to find out why now.

[innerlee]

Try to modify the cmd format first

[hellock]

This line is too weird. The following is more straight forward.

MMDET_SHARED_HEADS.register_module(ResNet3dLayer)

[kennymckormick]

However, registry.register_module() return a function,

@MMDET_SHARED_HEADS.register_module()
class ResNet3dLayer():
    pass

is equivalent to:

MMDET_SHARED_HEADS.register_module()(ResNet3dLayer)

[codecov]

Codecov Report

Merging #495 (9f0bf1b) into master (54cddd7) will decrease coverage by 0.04%.
The diff coverage is 33.33%.

@@            Coverage Diff             @@
##           master     #495      +/-   ##
==========================================
- Coverage   84.71%   84.66%   -0.05%     
==========================================
  Files         120      120              
  Lines        8407     8414       +7     
  Branches     1373     1373              
==========================================
+ Hits         7122     7124       +2     
- Misses        936      942       +6     
+ Partials      349      348       -1     

Flag	Coverage Δ
unittests	84.65% <33.33%> (-0.05%)	⬇️

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

Impacted Files	Coverage Δ
...action/core/bbox/assigners/max_iou_assigner_ava.py	51.02% <33.33%> (-1.07%)	⬇️
mmaction/models/backbones/resnet3d.py	90.40% <33.33%> (-0.25%)	⬇️
mmaction/models/backbones/resnet3d_slowfast.py	84.21% <33.33%> (-0.50%)	⬇️
mmaction/models/backbones/resnet3d_slowonly.py	75.00% <33.33%> (-5.00%)	⬇️
mmaction/models/heads/bbox_head.py	80.32% <33.33%> (-0.67%)	⬇️
mmaction/models/heads/roi_head.py	28.26% <33.33%> (-0.63%)	⬇️
...maction/models/roi_extractors/single_straight3d.py	83.33% <33.33%> (-1.58%)	⬇️
mmaction/core/evaluation/accuracy.py	93.18% <0.00%> (+0.90%)	⬆️

---

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 54cddd7...9f0bf1b. Read the comment docs.

[innerlee]

CI has been fixed in master, so you can remove changes in build.yaml

[kennymckormick]

CI has been fixed in master, so you can remove changes in build.yaml

Maybe you can still merge my modification. Since there is no built wheel for torch1.5.1 (which means you still can not pass the GPU CI), I modify build.yml to run CI under torch1.5.0 + torchvision 0.6.0.

[kennymckormick]

CI has been fixed in master, so you can remove changes in build.yaml

Seems although https://download.openmmlab.com/mmcv/dist/cu101/torch1.5.1/index.html is an empty page, you can still install mmcv-full and pass the GPU CI. I'm not sure how does that happen, would you like to share the explanation with me?

[innerlee]

🤷 have no idea.
Ask @ycxioooong

[ycxioooong]

@kennymckormick There is no built wheel for torch1.5.1 from a long time ago. We may have a discussion on this issue offline.

[dreamerlin]

How is it going ? @kennymckormick

[kennymckormick]

How is it going ? @kennymckormick

I think OK to merge.