Hi, Thank you for your code.
I have question about the nonsense situation.

As you know, visualizer.display_current_results save function saves the result images. Maybe the result from generator 'G' outputs with latest weights learned.
However, the synthesized image from the inference mode by test.py, is being terrible even though train-set and test-set are same. It makes nonsense. Because I saw already the G can makes good synthesized images from same dataset. While 50~100 epoch, I can see always good synthesized images. And the weight file should get same result in inference mode. is it right?

is that anything that I missed?

Thank you.