AliaksandrSiarohin

You can just use a single frame video, this will be effectively an image base swap.

Here is the [checkpoint](https://yadi.sk/d/6vqE41BTZKIAKQ), I did not compute all the metrics for this one. But it should be similar to the ones reported in the paper.

What data do you use, as source image and as a driving video?

The model is only trained on Taichi, it will not generaliza outside the dataset. You need Taichi guy for source image.

Hi, The difference between the vox-full.yaml and vox.yaml is the fact that vox.yaml run keypoint detector and dense motion on downscaled 64x64 images. On the other hand vox-full train all...

No. It is https://github.com/AliaksandrSiarohin/first-order-model But I can not provide a checkpoints yet, because of some privacy related issues.

Probably they are here https://drive.google.com/file/d/1IJ6sU5ynTKga4YdaG9PaW7lz-Zn4BBjh/view?usp=sharing

Hi, it doesn't matter, since optimizer for generator only updates generator parameters, while optimizer for discriminator only update discriminator parameters.

No didn't have it. Have you specified firtst order model flag?

And what is your torch version?