Code: github.com/PSRahul/dcgan
A deep convolutional GAN was trained from scratch on the Large-scale CelebFaces (CelebA) dataset of 200k face images.
A random sample vector as well as dropout was used in both the discriminator and the generator to induce noise. Various dropout values were tried.
Training
Download the images into the data/celeba/ folder in the same directory, then run:
./train.py --d1 [argument] --d2 [argument]d1 and d2 are dropout values applied to the odd and even layers of the network respectively. They can be chosen independently.
Sample generation
./sample_generator.py --d1 [argument] --d2 [argument]State dictionaries for trained models with different dropout values are provided in the models folder.
Sample results
