Question:
I am currently training a Generative Adversarial Network (GAN) for the task of image super-resolution. However, I am encountering a persistent issue where the generator’s loss decreases significantly over time, but the discriminator’s loss remains high and does not show any signs of improvement. This imbalance in training progress is preventing the discriminator from effectively distinguishing between real and generated high-resolution images.