[CVPR 2020] MERL-RAV Dataset contains over 19k faces annotated with 68 landmarks, with the additional information of whether each landmark is unoccluded, self-occluded or externally occluded.
Thanks for the amazing work. I have tried to use the loss in the paper on the facial landmark project. However , I find the loss cannot convergence. In chapter 3.2.1 of the paper, if the predicted landmark's value equal to the target, the loss's T1 part will tends to negative infinity.(Σij will tends to zero).Could you please tell me the correct way to use the loss? ^_^