Add ShuffleNet v2

[barrh]

Added 4 configurations: x0.5, x1, x1.5, x2
Added 2 pretrained models: x0.5, x1
Tested with Imagenet 1k.

Partially addresses: #645

TODO: re-upload the models from my branch, and update the URLs

[codecov-io]

Codecov Report

Merging #849 into master will increase coverage by 1.3%.
The diff coverage is 84.04%.

@@           Coverage Diff            @@
##           master    #849     +/-   ##
========================================
+ Coverage      54%   55.3%   +1.3%     
========================================
  Files          36      37      +1     
  Lines        3346    3401     +55     
  Branches      549     551      +2     
========================================
+ Hits         1807    1881     +74     
+ Misses       1406    1385     -21     
- Partials      133     135      +2

Impacted Files	Coverage Δ
torchvision/models/__init__.py	100% <100%> (ø)	⬆️
torchvision/models/shufflenetv2.py	83.87% <83.87%> (ø)
torchvision/datasets/caltech.py	19% <0%> (-2.14%)	⬇️
torchvision/datasets/sbd.py	31.66% <0%> (+0.08%)	⬆️
torchvision/models/resnet.py	83.55% <0%> (+0.21%)	⬆️
torchvision/datasets/celeba.py	16.09% <0%> (+1.35%)	⬆️
torchvision/datasets/imagenet.py	22.68% <0%> (+1.47%)	⬆️

---

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 8759f30...4c8abb6. Read the comment docs.

[fmassa]

This looks generally very good, thanks!

I've a question regarding the pre-trained model: what were the hyperparameters / how did you train them? Is the codebase available somewhere?

[barrh]

I didn't train the pretrained models. The weights were published here: https://github.com/ericsun99/Shufflenet-v2-Pytorch

[barrh]

@fmassa is there any other change request, or can we go on to merge this addition?

[fmassa]

LGTM, thanks!

I'll update the weights path in a later commit.