Loss does not go down and explodes eventually!

Open slymnyldrm opened this issue 2 years ago • 1 comments

Hello, I am trying to train your model on a single A6000 from scratch with DF2K dataset. First, I have tried your original config and then I have tried batch_size=18 and LR_size=64. Nevertheless, I couldn't train the model. The loss remains at 1 and eventually explodes. Output images is just random rgb noise. Would you be able to help me with this problem?

Thank you

Aug 06 '23 10:08 slymnyldrm

There should be some mistakes in your config. The most possible reason is that you did not load the Autoencoder model correctly.

Aug 31 '23 18:08 IceClear