Generative Adversarial Networks(GAN) in TensorFlow 2.0

Tensorflow implementation of deep learning pipelines that are based on Generative Adversarial Networks. Now it type of GAN is available such as:

• Vanilla GAN (Mnist)
• Vanilla GAN (Fashion Mnist)
• DCGAN (Planned)
• StyleGAN (Planned)

Prerequisites

Usage Example - Vanilla GAN for MNIST

Pre-trained models To test with an existing model:

$ python3 run.py --exp=PRETRAINED  --problem=VANILLA_MNIST

Training models Or you can train by yourself

$ python3 run.py --exp=TRAIN  --problem=VANILLA_MNIST

Results

Due to time reasons, the following models have a low number of learning epochs.

Image generation

Vanilla GAN (MNIST)	Vanilla GAN (FASHION_MNIST)

Loss

Vanilla GAN (MNIST)	Vanilla GAN (FASHION_MNIST)

Monitoring model training

...WIP...

References

1. Generative Adversarial Networks
2. Deep Convolutional Generative Adversarial Network Tutorial in TensorFlow