Python notebook containing TensorFlow DCGAN implementation. It was trained on a Simpsons Faces dataset.
Check out corresponding Kaggle kernel: Image Generator.
Check out corresponding Medium article:
Image Generator - Drawing Cartoons with Generative Adversarial Networks
Network architecture by Radford et al., 2015.
Visualization of training with the following hyperparameteres.
IMAGE_SIZE = 128
NOISE_SIZE = 100
LR_D = 0.00004
LR_G = 0.0004
BATCH_SIZE = 64
EPOCHS = 300
BETA1 = 0.5
WEIGHT_INIT_STDDEV = 0.02
EPSILON = 0.00005
Cherry-picked generated samples.
As expected, there were some funny-looking malformed faces as well.
Greg (Grzegorz) Surma