[Feature] enable exporting to onnx for PointRend

[DmitriySidnev]

This PR enables exporting to ONNX for PointRend based models.

[CLAassistant]

All committers have signed the CLA.

[codecov]

Codecov Report

Merging #953 (84cd692) into master (d212bd5) will decrease coverage by 0.15%.
The diff coverage is 54.21%.

❗ Current head 84cd692 differs from pull request most recent head 152aa19. Consider uploading reports for the commit 152aa19 to get more accurate results

@@            Coverage Diff             @@
##           master     #953      +/-   ##
==========================================
- Coverage   65.72%   65.56%   -0.16%     
==========================================
  Files         157      154       -3     
  Lines       10092     9950     -142     
  Branches     1828     1806      -22     
==========================================
- Hits         6633     6524     -109     
+ Misses       3113     3090      -23     
+ Partials      346      336      -10     

Flag	Coverage Δ
unittests	65.56% <54.21%> (-0.16%)	⬇️

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

Impacted Files	Coverage Δ
mmcv/ops/point_sample.py	44.02% <54.21%> (+24.02%)	⬆️
mmcv/tensorrt/__init__.py	50.00% <0.00%> (-10.00%)	⬇️
mmcv/tensorrt/init_plugins.py	43.75% <0.00%> (-6.25%)	⬇️
mmcv/ops/nms.py	56.42% <0.00%> (-5.21%)	⬇️
mmcv/runner/hooks/logger/base.py	68.93% <0.00%> (-1.95%)	⬇️
mmcv/runner/hooks/logger/wandb.py	68.75% <0.00%> (-0.95%)	⬇️
mmcv/runner/hooks/logger/mlflow.py	80.64% <0.00%> (-0.61%)	⬇️
mmcv/runner/hooks/logger/pavi.py	70.58% <0.00%> (-0.43%)	⬇️
mmcv/ops/deform_conv.py	61.76% <0.00%> (-0.28%)	⬇️
mmcv/cnn/__init__.py	100.00% <0.00%> (ø)
... and 19 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 d212bd5...152aa19. Read the comment docs.

[RunningLeon]

@DmitriySidnev To fix lint, could you merge master and push again?

[zhouzaida]

hi @DmitriySidnev , CI failed.

[DmitriySidnev]

@zhouzaida, hi! I have merged master branch. Hope it helps. With onnxruntime==1.5.1 all onnx tests passed successfully.

[zhouzaida]

@zhouzaida, hi! I have merged master branch. Hope it helps. With onnxruntime==1.5.1 all onnx tests passed successfully.

got it

[DmitriySidnev]

In the failed job build / build_cpu (3.7, 1.7.0) (pull_request) I can see the following lines:

Collecting onnxruntime==1.4.0
  Downloading onnxruntime-1.4.0-cp37-cp37m-manylinux2010_x86_64.whl (4.4 MB)

But this version of onnxruntime does not satisfy this statement.

[RunningLeon]

@DmitriySidnev You could find the reason why test_bilinear_grid_sample failed with torch==1.7.0 on cpu.

[DmitriySidnev]

@RunningLeon, bilinear_grid_sample function fixed. In onnxruntime 1.5.1 onnx models obtained from pytorch 1.7.0 and 1.8.1 now work.

[RunningLeon]

@DmitriySidnev Thanks for the effort. Could you fix CI error? Thanks again.

[DmitriySidnev]

@RunningLeon, the bilinear_grid_sample function fixed for pytorch < 1.8.

[RunningLeon]

kindly ping @ZwwWayne

[RunningLeon]

LGTM. Kindly ping @ZwwWayne

[RangiLyu]

LGTM