search

google / compare_gan

Compare GAN code.
Apache License 2.0
1.82k stars 317 forks source link

resnet_biggan_deep seems not consist with BigGanDeep in the paper #34

Closed RuiLiFeng closed 5 years ago

RuiLiFeng commented 5 years ago

In line 260, resnet_biggan_deep.py, z already be set into tf.concat([z,y],1), while in line 285, when feed parameters to the resnet block, z is treated as the original latents code and y is repeatedly feed into the block, which means that the resnet block in fact get y twice. This seems inconsistent with the settings in the biggan paper. image image image

Marvin182 commented 5 years ago

In this case z and y are indeed the same but only one is used. We pass around both for consistency with other architectures. Since the only the part in the ResNet block that uses uses z or y is the batch norm, you can configure any of the batch norms and the code should run. I don't think we have batch norm implementation that use both z and y.