Closed cscyangj closed 5 years ago
Hi! So we were given the crops and just resized them to 256x256. As you can see it's rarely not square by more than 1/2 pixels.
I'm going to go ahead and close the issue. (We need to resize for training -- the network needs a 256x256 input. otherwise we'd have to pad the image.)
Hi,
It is said that When training the identity, video dataset using dlib to crop the face to 256x256, as i can see from the example image, size is not exactly 256x256.
So the reason for adjust image size is for better result or other reasons?