nnuyi/sngan

A implement of spectral normalization GAN for tensorflow version

sngan's Issues

同学，能解释下update_collection是什么意思啊？

我在一个韩国人复现的SNGAN里面没有看到update_collection。

tf.matmul（a,b）我不知道等不等于tf.matmul（b,a）。
反正你们两个人的复现tf.matmul都与作者给的流程图算法里面的运算位置相反

初学者的问题

同学你好，我是深度学习的初学者。我想问下，把你所给输入图像大小改成256*256，输出图像也是同样大小。卷积核大小、个数需要改变吗？或者说程序中需要改动哪里？谢谢了

Could it be possible to make the trained GAN publicly available?

I have tried to train the GAN on my personal computer (with cheap GPU), but it takes too long. I think having these trained GAN publicly available would be really helpful for people that cannot afford to have big GPU clusters. Thank you in advance!

WGAN-GP加SN，两个loss都是nan

同学实在是不好意思，叒问你问题。
我把你写的SN加到WGAN的判别器中，生成和对抗loss值都是nan。
我把WGAN-GP的梯度惩罚项去掉，生成和对抗loss值不是nan。
这是什么原因啊？讲道理，SN是对判别器起作用，为什么会对生成器loss产生影响？
再次感谢！

How to use in 3D conv?

The paper and the code are both for the 2D convolution of the sn limit w, then how to deal with w in the 3D convolution?

Why do you use `update collection`?

While skimming through your code, I don't understand why do you need update_collection variable here,
https://github.com/nnUyi/SNGAN/blob/master/SNGAN.py#L125
Later on I go deeper but don't understand here what do you want to do bt the following section of code,
https://github.com/nnUyi/SNGAN/blob/master/ops.py#L91-L97