Some trouble when using 128x128 shape of image. about psfrgan

[chaofengc]

The model is a simple U-Net and should work with 128x128 inputs if you retrain it.

It seems that there are some problems in your code when loading the retrained parsing network. Possibly because you save the network without DataParallel while the default load_weight function expect a model wrapped with DataParallel.

I am not sure what is the exact problem. Sorry that I am not supposed to help you with the problem, you may use some debug tools like pdb to fix the bugs by yourself.

[fbc-one]

Thank you! I have successfully solved the problem. It was indeed caused by the mismatch of the network name saved in the pre-trained model. Currently, I am able to generate mask images with a shape of 128. Sincerely appreciate your reply!