DCGAN Simpsons Faces Image Generator usging Tensorflow2 Alpha 🖼️.

Reimplemented Greg's image generator as tensorflow2-alpha.
Greg's image generator can be found here.
The reimplementation used Greg's hyperparameters and datasets.

Python notebook containing TensorFlow2 DCGAN implementation. The data was downloaded as described by Greg. It was trained on a Simpsons Faces dataset.

I applied the list of image files that Greg excluded. ( simpsons_dataset.txt )

Check out corresponding Kaggle kernel: Image Generator.

Greg result :

Network architecture by Radford et al., 2015.

The hyperparameteres applied by Greg were used. 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

Selected Samples

I also had some funny-looking malformed faces as well

JiamSeo (jams777) Donny

GITHUB