First of all, I would like to congratulate you on your achievements and thank you for exposing the code to provide quick learning for future generations.
After the analysis, I have three questions in model 2: (inpainting part)
EVAL_INTERVAL in config is 0, so did you not add validation to your training?
When testing, new input and mask are provided, so the test set of the original celeba data is not used?
In the model 2 training section, the training will not stop until 1999999 iterations, which is obviously over-fitting. How to get a good inpaiting model?
Thank you
First of all, I would like to congratulate you on your achievements and thank you for exposing the code to provide quick learning for future generations. After the analysis, I have three questions in model 2: (inpainting part)