Keys for mvit model and its pretrained weights are not matched.

[jaelgu]

Fail to load pretrained mvit model:

import torch

model = torch.hub.load('facebookresearch/pytorchvideo', model='mvit_base_16x4', pretrained=True)

• it works with my previous copy of ~/.cache/torch/hub/facebookresearch_pytorchvideo_main/
• the issue happens when I use the latest pytorchvideo

[Anirudh257]

Try this: #164

[jaelgu]

#164

Still the same error

[jaelgu]

Some workaround: replace the pytorchvideo in torch cache with some previous version.

I tried with tag v0.1.3 and it works. However, better it can be fixed in latest code as well.

[Anirudh257]

Okay, I think that they forgot to update the code. Try with an older commit of the repo too. Like I did here: facebookresearch/SlowFast#456 (comment)