buriburisuri / supervised_infogan Goto Github PK

Hi.

Since TensorFlow API has been chanced, such as

# legacy: tf.concat(concat_dim, values)
tf.concat(values, axis) 

so this program return below error messages.

TypeError: Expected int32, got list containing Tensor of type '_Message' instead.

So, if you use >= 1.0 version of TensorFlow, you should change concat code.
For example,

# train.py line 69: 
# legacy: xx = tf.concat(0, [x, gen])
xx = tf.concat([x, gen], axis=0)

As far as I know, GAN is designed for unsupervised learning. Is the use of the label information making GAN meaningless?
Thanks.

when I run train.py, it shows the error message: "TypeError: Expected int32, got list containing Tensors of type '_Message' instead."
I'm wonder what's the problem is?
I didn't modify any code, just run the downloaded package.

Hi, @buriburisuri Thank you very much with this project, it is very interesting!!!
However, I have a problem to understand this line here generator loss. Could you explain the disc.sg_reuse(input=gen).sg_bce(target=y) .
I have check your code in sugartensor, but it seems to be too hard to understand

Thank you again for sharing this project