On the problem of stagnation of training ESRGAN generator

[zkh123456789]

[zkh123456789]

HI xintao ，when I used my own dataset to train the ESRGAN generator, I found that there had been no response for a long time. What is the cause of this? The second picture shows that the code is running, taking up the memory of GPU.

[zkh123456789]

[zkh123456789]

@xinntao

[DejavuSans]

@zkh123456789 hi, you can add some print log in code to check whether the code is running as you expected. If you use "disk" type for the io_backend in .yml file, the code will traversal the whole dataset, this will take a long time. Of course, you can use a very small dataset to debug the training stage, after that you can train on a large dataset.