I have trained a GAN on Imagenet dataset which performs the task of filling the missing regions of an image by considering both L2 loss and Adverserial Loss using Pytorch.
As a part of the testing process I loaded the above generator model and performed a forward pass of the test image through the generator network to fill the missing region.
To my surprise the missing region was filled by random noise. Can I know what might be the reasons:
1.)Is this the case of overfitting? Given the fact that I did try using the saved generator model on one of the trained images, and it still ends up filling the missing region with random noise.
2.)Or is it a problem in the way I am loading the model?