[ONNX] fix dynamic resize

[neginraoof]

[ONNX] fix dynamic resize

[codecov]

Codecov Report

Merging #2488 into master will increase coverage by 1.83%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master    #2488      +/-   ##
==========================================
+ Coverage   68.76%   70.60%   +1.83%     
==========================================
  Files          94       94              
  Lines        7995     7995              
  Branches     1272     1272              
==========================================
+ Hits         5498     5645     +147     
+ Misses       2088     1947     -141     
+ Partials      409      403       -6     

Impacted Files	Coverage Δ
torchvision/models/detection/transform.py	95.55% <ø> (+17.22%)	⬆️
torchvision/io/video.py	69.82% <0.00%> (+1.18%)	⬆️
torchvision/models/detection/rpn.py	93.13% <0.00%> (+2.57%)	⬆️
torchvision/models/detection/roi_heads.py	82.34% <0.00%> (+5.10%)	⬆️
torchvision/__init__.py	78.12% <0.00%> (+6.25%)	⬆️
torchvision/datasets/video_utils.py	75.75% <0.00%> (+7.27%)	⬆️
torchvision/ops/boxes.py	84.44% <0.00%> (+11.11%)	⬆️
torchvision/ops/poolers.py	97.02% <0.00%> (+11.88%)	⬆️
torchvision/io/_video_opt.py	47.50% <0.00%> (+26.25%)	⬆️
... and 1 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 d481f2d...32acc81. Read the comment docs.

[neginraoof]

@francisco Vitor Suzano Massa

Could you let us know if tests from vision release/0.7 has been tested against pytorch release/1.6 branch?
I believe the tests would fail due to updates in recompute_scale_factor. Here is a quick patch we can have just to unblock release: #2488

[fmassa]

Thanks!