search

zhangqianhui / progressive_growing_of_gans_tensorflow

Tensorflow implementation of PROGRESSIVE GROWING OF GANS FOR IMPROVED QUALITY, STABILITY, AND VARIATION
MIT License
149 stars 71 forks source link

One Pix_Norm too much #8

Open A-Specker opened 6 years ago

A-Specker commented 6 years ago

but we perform pixelwise normalization of the feature vectors after each Conv 3 × 3 layer in the generator as described in Section 4.2.

They write this in A.1 in the paper.

Meaning the line 265 inPCGAN.py is not intended in the original.

zhangqianhui commented 6 years ago

thank you , Can you make a PR?

A-Specker commented 6 years ago

Hmm.. I didn't even clone it, so that would exceed my small finding

zhangqianhui commented 6 years ago

ok

zhangqianhui commented 6 years ago

I see the official code, they use pixelnorm layer for 4x4 kernel conv.

    net = ReshapeLayer(name='Ginb', incoming=net, shape=[[0], [1], 1, 1])
    net = PN(BN(WS(Conv2DLayer(net, name='G1a', num_filters=nf(1), filter_size=4, pad='full', nonlinearity=act, W=iact))))
    net = PN(BN(WS(Conv2DLayer(net, name='G1b', num_filters=nf(1), filter_size=3, pad=1,      nonlinearity=act, W=iact))))
lods = [net]
zhangqianhui commented 6 years ago

https://github.com/tkarras/progressive_growing_of_gans/blob/master/network.py

zhangqianhui commented 6 years ago

@A-Specker

A-Specker commented 6 years ago

Ok, so Issue solved, I just read their paper, not the code.