dual_contrast_loss argument - error while training

[TheodoreGalanos]

When trying to train a model, the dual_contrast_loss argument spits up an error on lightweight-gan.py. My guess is that it's not implemented yet.

At first I thought it was a mismatch (should it be dual_contrastive_loss btw like in lightweight-gan?) but apparently needed to be commented out for training to start successfully.

Very exciting by the way! Training seems to be super fast, about 28h for 150,000 iterations for a 256p model on my 2080 (although it's a very small dataset). I'll share some closer-to-final results once I have them.

Best, Theodore.

[Iceland-Leo]

Looking forward to your closer-to-final results !

[Devetec]

@TheodoreGalanos Are you talking about transganformer or lightweight GAN?