Amount of training steps about lightweight-gan

[Dok11]

I have been 550 000+ steps for my images and result still not enough good.
But in fact the result was not improve (subjective, I didnt calc FID) after 30-60k steps.

[MartimQS]

Have you tried logging the values for the loss functions and gp in a file and visualizing the results?

[Dok11]

The loss values is not informative and dont have relative to visual quality like FID, but FID require other installed library and a lot of time so I still not calulate this =)
Visualising result is enabled by default, and I made command to show progress (see readme.md) and I really dont see improve after 60k steps (with batch size equal to 3)

[MartimQS]

The visualization of the result shows the generated images. However, I’m still improving the training phase and discovering the best settings combination for training, that is why I wanted to know the behavior of the networks with each iteration