AliaksandrSiarohin

Hi, sorry but your questions is really confusing: 1. I don't get the question. There is no L1 in train mode. 2. What is 768? 3. Could you provide example?...

1. Reconstruction does not make sense for avd, since it specifically designed for cross identity, where the shapes of the objects could be different. 2. There are no explicit handling...

You can try Ted-YouTube model.

It is trained on more data.

We did not include vox, because there is no benifit in using current model. As explained in the paper. The checkpoint and config for vox is still provided.

Just a speculation. This model has better generalisation when estimating affine transformations, you can see explanation for this phenomenon in the toy experiment section. Because we evaluate model on faces...

You should train until the very end.

You should update YouTube-dl.

Can you download it using command line? Can you access these videos from browser?

@Linda0530 I don't get it do you still have a problem? Did you see #12?