Architecture of Generator Network.

[kvthr]

In the code, it says CONV->BIAS->ACTIVATION but it is CONV->BATCH NORMALISATION->ACTIVATION. Correct me if I am wrong!

[blaskowitz100]

Hey,
I think the batch normalization already includes the bias term. See here:
https://stackoverflow.com/questions/46256747/can-not-use-both-bias-and-batch-normalization-in-convolution-layers

[patrikohlsson]

Actually bias is built in to tf.layers.conv2d and tf.layers.conv2d_transpose and is on by default. So technically it's conv->bias->batch normalization->activation

[patrikohlsson]

But as @blaskowitz100 mentioned it's common to scale and add a bias after batch normalization as well.