search

kozistr / ESRGAN-tensorflow

Enhanced SRGAN. Champion PIRM Challenge on Perceptual Super-Resolution
https://arxiv.org/abs/1809.00219
Apache License 2.0
34 stars 5 forks source link

Error in loss #4

Open ChristofferEdlund opened 5 years ago

ChristofferEdlund commented 5 years ago

https://github.com/kozistr/ESRGAN-tensorflow/blob/b980f5e856e663be4115a6883c69d8c9b7e66b8e/losses.py#L8-L11

            real = real - tf.reduce_mean(fake)
            fake = fake - tf.reduce_mean(real)

I presume that you want the original real variable and not the relativistic one when calculating fake?