A DCGAN to generate Pokemon. The dataset I have used has ~2800 images shrunk down to (100, 100) from the original size, and each model has been trained over 50 epochs.
(I have used a smaller dataset as GitHub has a file size limit of 100MB)
The original dataset that consists of ~11,000 images can be found here
Here are a few images generated across a couple of models.
-
- Learning rate 1e-4
- Adam optimizer for both generator and discriminator
-
- Learning rate 1e-4
- Adam optimizer for generator and SGD optimizer for discriminator
-
- Learning rate 1e-4
- Adam optimizer with momentum = 0.5 for generator
- SGD optimizer for discriminator
All images shown above can be found here
The graphs for the different models over 50 epochs can be found here
Code is in the src folder