Brandon Xie

It seems that the dataset released contains only the annotation of 11 facial parts. However, the repository also provide the model trained with dataset containing labels of 14 facial parts....

confusion concerning the kl loss

5

The output of fc1 can be negative while traning. And in KL loss computing procedure, negative value is not valid. So I wonder how will you explain or modify it.

Missing Videos in Three Datasets (Vox, Taichi and Ted)

3

Thanks for your efforts in building this project. I am so grateful for it. However, we found that some of the videos in three datasets(Vox, Taichi and Ted) can't be...