search

duxingren14 / DualGAN

DualGAN-tensorflow: tensorflow implementation of DualGAN
Apache License 2.0
265 stars 97 forks source link

Is batch normalization still used in training mode when sampling examples (testing) ? #10

Closed watsonyanghx closed 4 years ago

watsonyanghx commented 7 years ago

I see you don't specify the running mode in batch_norm. Is there any reasons? Does this have a big influence in the quality of image generated?

duxingren14 commented 7 years ago

Thanks for pointing out this. Actually I used the default setting of BatchNorm for all my experiments. No further trials indicate if a different running model does bring any influence. Anyway it is an interesting topic that deserves a trial.

watsonyanghx commented 7 years ago

Thanks for your reply.

I have tried to split train and test phase, and the images sampled in training mode looks good, but when changed to testing mode (set is_training=False of batch normalization), all the images generated looks very bad， they are just noisy points.