Models not pushed to torch hub

oke-aditya commented 3 years ago

📚 Documentation

Hi!

The following models are listed in our hubconf.py but these models are not listed over pytorch hub

[ ] MobileNetv3: - Mobilenet v3 small, Mobilenet v3 large
[ ] Mobilenet Segmentation: - lraspp_mobilenet_v3_large
[x] DeepLab Models: - deeplabv3_mobilenet_v3_large deeplabv3_resnet50
[x] FCN: - fcn_resnet50,
[ ] MNasNet: - mnasnet0_5, mnasnet0_75, mnasnet1_0, mnasnet1_3
[ ] EfficientNet: - efficientnet_b0, efficientnet_b1, efficientnet_b2, efficientnet_b3, efficientnet_b4, efficientnet_b5, efficientnet_b6, efficientnet_b7

Additional Context: -

I think there is ongoing issue with torch hub. #4156 Also we do not publish detection models to torch hub see #1945

P.S. Please verify the list and feel to free to edit if I have missed something. Also probably this issue can be transferred to torch hub repository.

cc @datumbox @NicolasHug

datumbox commented 3 years ago

Thanks for bringing this up @oke-aditya. I think it's reasonable to include them.

The only concern I got is for mnasnet0_75 and mnasnet1_3 which don't have pre-trained weights, but all other look good to me. Would you like to send a PR?

oke-aditya commented 3 years ago

Yes, I will! Happy to help :smile:

NicolasHug commented 3 years ago

Thanks @oke-aditya ,

as a minor caveat we should just make sure that those models aren't relying on custom ops, which is the reason why the detection models aren't published yet: https://github.com/pytorch/vision/issues/1945#issuecomment-597062931

(I haven't checked myself)

datumbox commented 3 years ago

All models listed here don't require require custom operators that need compilation. He should be good to go.

pytorch / vision

Models not pushed to torch hub #4289

📚 Documentation