[WIP] Add commands for model training

[fmassa]

This adds a README explaining the commands that have been used to train particular models.

I've started with the latest classification models that I've trained for the 0.3 release of torchvision.
I'll improve the scope of it to cover all models in torchvision, which will probably involve retraining some old models to get reference hyperparameters.

cc @andravin

[codecov-io]

Codecov Report

Merging #1203 into master will increase coverage by 0.05%.
The diff coverage is n/a.

@@            Coverage Diff            @@
##           master   #1203      +/-   ##
=========================================
+ Coverage   65.55%   65.6%   +0.05%     
=========================================
  Files          74      74              
  Lines        5780    5780              
  Branches      883     883              
=========================================
+ Hits         3789    3792       +3     
+ Misses       1726    1724       -2     
+ Partials      265     264       -1

Impacted Files	Coverage Δ
torchvision/transforms/transforms.py	80.94% <0%> (-0.2%)	⬇️
torchvision/transforms/functional.py	71.38% <0%> (+1.15%)	⬆️

---

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 7253e81...ce3df33. Read the comment docs.