search

osh / KerasGAN

A couple of simple GANs in Keras
501 stars 177 forks source link

Inconsistent Shape of Generator output #4

Open malzantot opened 7 years ago

malzantot commented 7 years ago

Hi,

Thanks for the example. I tried running it using Keras v1.1.1 , However the code generates an an inconsistent shape of the generator output. The output I get is [None, 400, 28, 1] instead of [None, 1, 28, 28] Therefore, it fails to stack the two models together.

Problem happens with both Theano and Tensorflow backends. any hints ?

malzantot commented 7 years ago

Seems like the issues comes from the Keras Convolution2D which requires to specify which dimension of input image stands for the depth. Problem solved by adding the following , dim_ordering='th' to the Convolution2D and Sampling2D layers.

malzantot commented 7 years ago

Apparently the problem is still there. Although it could successfully stack the two models after I added the dim_ordering flag. It now fails when I try to run the generative model.

It gives me the following error message ValueError: CorrMM images and kernel must have the same stack size