How to interpret the following loss graph of CGAN?

I am training CGAN for generating Xray images.

My training initializations are as follows:

criterion = nn.BCEWithLogitsLoss()
training samples = 1200
n_epochs = 600
z_dim = 100
batch_size = 50
lr_gen = 0.0002
lr_disc = 0.0001

CGAN-loss graph

This is the loss graph after 500 epochs. I am not able to interpret whether training is working fine or not. What does this generator’s and discriminator’s graph conclude?

And following is the output of fake images for 554th epoch.

The losses would indicate that the discriminator is successful in differentiating fake from real examples (thus the discriminator loss is low) while the generator has trouble creating images which the discriminator would classify as “real” images.

1 Like