The result of 20W training is not ideal

[Funple]

The result of 20W training is not ideal. Is it because I didn't pay attention to the key points in training? And the results of using the pre-train model you provided are not ideal. I observed that in tensorboard distribution, there was always a floating and no downward trend.

[Funple]

十分感谢能够使用您如此规范的代码，期待您的回复。

[rkfg]

That's a common thing for GANs it seems, if the background becomes negative you have to restart training.

[Cosica]

Hello, could you teach me how to use the pretrained model? I entered "python inference.py --model pretrained/horse2zebra.pb --input data/testA --output output" in my Pycharm's terminal but it shows "tensorflow.python.framework.errors_impl.UnknownError: NewRandomAccessFile failed to Create/Open: data/testA : \udcbe\u073e\udcf8\udcb7\udcc3\udcce\u02a1\udca3
; Input/output error

"