neural-head-avatars/deps/RobustVideoMatting/model/decoder.py", line 169, in forward_single_frame r, z = self.ih(torch.cat([x, h], dim=1)).split(self.channels, dim=1) RuntimeError: Sizes of tensors must match except in dimension 1. Expected size 8 but got size 18 for tensor number 1 in the list.

Question

neural-head-avatars/deps/RobustVideoMatting/model/decoder.py", line 169, in forward_single_frame r, z = self.ih(torch.cat([x, h], dim=1)).split(self.channels, dim=1) RuntimeError: Sizes of tensors must match except in dimension 1. Expected size 8 but got size 18 for tensor number 1 in the list.

Closed this issue 2 years ago · 4 comments

neural-head-avatars/deps/RobustVideoMatting/model/decoder.py", line 169, in forward_single_frame
r, z = self.ih(torch.cat([x, h], dim=1)).split(self.channels, dim=1)
RuntimeError: Sizes of tensors must match except in dimension 1. Expected size 8 but got size 18 for tensor number 1 in the list.

Answer 1 · 2022-04-25T15:30:41.000Z

I've tried to reproduce this but it doesn't show up for different videos I tested. Providing more details on how you encountered it would be helpful.

Answer 2 · 2022-04-26T02:27:35.000Z

@philgras it might be happend when somehow 2 heads detected. There might no such handling in code.

Answer 3 · 2022-04-26T16:18:43.000Z

Yes, we don't handle that case. Also mind that it is important that the subject is visible in every frame as the tracking algorithm and the avatar optimization rely on valid preprocessing outputs for each frame. Feel free to close this issue, if you think the problem is clarified.

Answer 4 · 2022-04-27T02:55:23.000Z

@philgras it is. I would suggest if we can handle it or put some warnings to reminder user would be more friendly for newcomers. thanks for your clarification though.