Awesome
BigGAN-Tensorflow
Simple Tensorflow implementation of "Large Scale GAN Training for High Fidelity Natural Image Synthesis" (BigGAN)
Issue
- The paper used
orthogonal initialization
, butI used random normal initialization.
The reason is, when using the orthogonal initialization, it did not train properly. - I have applied a hierarchical latent space, but not a class embeddedding.
Usage
dataset
mnist
andcifar10
are used inside keras- For
your dataset
, put images like this:
├── dataset
└── YOUR_DATASET_NAME
├── xxx.jpg (name, format doesn't matter)
├── yyy.png
└── ...
train
- python main.py --phase train --dataset celebA-HQ --gan_type hinge
test
- python main.py --phase test --dataset celebA-HQ --gan_type hinge
Architecture
<img src = './assets/architecture.png' width = '600px'>128x128
<img src = './assets/128.png' width = '600px'>256x256
<img src = './assets/256.png' width = '600px'>512x512
<img src = './assets/512.png' width = '600px'>Author
Junho Kim