Question about the loss function

vyom1911 / Facial-Landmark-Detection

This is an implementation of Task Constrained Convolution Neural Network (TCDCN) which detects 5 landmarks (Both eyes, nose and both ends of lips) on a facial picture. Our model take in a list of directory location of images and produces images with marked landmark in the output directory.

7 stars 2 forks source link

Question about the loss function #2

Open LeeBC2298 opened 4 years ago

LeeBC2298 commented 4 years ago

I think the facial landmark regression loss and other losses should add together, In the paper there is a subtraction because softmax cross entropy losses are negative.

jinneigeyu commented 3 years ago

I agree. and the papper says "when the gender , smile , glass , pose branch model are trainning good, these branches' parmas don't need to train ". These part wan not shown in this implementation