Training code for the networks described in "How far are we from solving the 2D & 3D Face Alignment problem? (and a dataset of 230,000 3D facial landmarks)" paper.
Hi, I am a newcomer of face alignment work and trying to train your model on my own data set. I read about hourglass model and find it is trained by intermediate supervision in previous work. I read your torch 7 training code (I am actually not familiar with torch 7) and still can not find the answer. Could your please tell me if you compute loss for each hourglass stage and do backprop or just for the final stage. Thx to your work!
Hi, I am a newcomer of face alignment work and trying to train your model on my own data set. I read about hourglass model and find it is trained by intermediate supervision in previous work. I read your torch 7 training code (I am actually not familiar with torch 7) and still can not find the answer. Could your please tell me if you compute loss for each hourglass stage and do backprop or just for the final stage. Thx to your work!