Weird results on example images about realistic-neural-talking-head-models HOT 4 CLOSED

Looks like you’ve used the video e_hat to train on the example image. After you run embedder_inference.py, make sure the path_to_embedding variable in finetuning_training.py is the actual path to the embedded image file. By default it should be e_hat_images.

from realistic-neural-talking-head-models.

@Bip3 thank you for your response. path_to_embedding = 'e_hat_images.tar' in my case.

from realistic-neural-talking-head-models.

Seems like doesn't work if I fine tune on the testing images and then feed landmarks from the testing example video during inference time. Probably both landmarks and identity image should be in the training data, or landmark should not be different too much from it.

from realistic-neural-talking-head-models.

I have met the same problem, while i use mark zuckerberg's video as landmark and testing example as embedding with pretrained model.

from realistic-neural-talking-head-models.