ZZUTK/Face-Aging-CAAE

Test image problem

okmmsky888 opened this issue · 2 comments

@ZZUTK Test the image, I used Dlib to face the face, and then cut. But when the prediction, face reconstruction deformation, fuzzy can not see, where is the problem?

ZZUTK commented

Maybe the training set is not diverse enough, or the trained model is not generative enough to generate reasonable images from a random z. Ideally, z should be uniformly distributed, which means any random sample from uniform distribution should yield a face. You may check the distribution of z, whether it is close to uniform noise.

@ZZUTK My training set of images 167988, size_batch set to 600, epoch = 50 This is no problem?