vdumoulin / morphing_faces Goto Github PK

Hi

Your code is nicely organized and well demonstrated. I'm just wondering if you could also post the code for training the VAE on github. I'm a bit confused about the reparametrization trick, how to sample Z and propagate the gradient during the training.

Thanks so much!

Hi,

I fed an arbitrary face image to the model by calling infer_Z and assigned the resulting code to the self.Z. then tried to generate the image using infer_Z(); however, the generated image is usually a black 48*48 page or with some random lines on it. Any suggestion to generate the input image successfully?
(All above modifications were made in morphing.py)

Thanks in advance

I read your python code and found out the decoding process is done by generating the face image directly from the code through the decoding network rather than by sampling from model. How does a face is sampled from a isotropic gaussian distribution with your code?