Refactor the get_weights API

[datumbox]

Fixes #4652

• Change the default weights mechanism to sue Enum aliases.
• Change get_weights to work with full Enum names and make it public.
• Test that all reference scripts continue to run.

cc @datumbox @bjuncek

[facebook-github-bot]

💊 CI failures summary and remediations

As of commit 6d0cc8e (more details on the Dr. CI page):

---

💚 💚 Looks good so far! There are no failures yet. 💚 💚

---

This comment was automatically generated by Dr. CI (expand for details).

Please report bugs/suggestions to the (internal) Dr. CI Users group.

Click here to manually regenerate this comment.

[datumbox]

The references were validated by running the following commands. All work as expected:

torchrun --nproc_per_node=1 train.py --test-only --weights ResNet50_Weights.ImageNet1K_V2 --model resnet50

python train_quantization.py --device cpu --test-only --weights ResNet50_QuantizedWeights.ImageNet1K_FBGEMM_V2 --model resnet50

torchrun --nproc_per_node=1 train.py --test-only --weights SSDLite320_MobileNet_V3_Large_Weights.Coco_V1 --model ssdlite320_mobilenet_v3_large

torchrun --nproc_per_node=1 train.py  --dataset coco --model lraspp_mobilenet_v3_large --test-only --weights LRASPP_MobileNet_V3_Large_Weights.CocoWithVocLabels_V1

torchrun --nproc_per_node=1 train.py --data-path /datasets01/kinetics/070618/ --train-dir=val_avi-480p --val-dir=val_avi-480p --batch-size=64 --sync-bn --test-only --weights R2Plus1D_18_Weights.Kinetics400_V1 --cache-dataset

[NicolasHug]

Thanks @datumbox , I made some minor comments and a question, but this LGTM

[NicolasHug]

I'm wondering: shoudn't this return a Weight instance, instead of WeightsEnum?
Same for from_str (and in the return section below).

[datumbox]

I think we need to return a WeightsEnum value. That is the value of an Enum which maintains information about the class it comes from (ResNet50_Weights). Returning a Weights loses the information necessary to validate that the right type of weights were passed to the method.