EBGAN : Energy-based Generative Adversarial Network

Overview

Pytorch implementation of EBGAN(arxiv:1609.03126)

Objectives

Architectures

Repelling Regularizer

Dependencies

python 3.6.4
pytorch 0.3.1.post2
visdom

Result : CelebA(aligned, 64x64)

(you can download CelebA dataset here)

python -m visdom.server
python main.py --dataset CelebA --epoch 15 --batch_size 32 --PT_ratio 1 --m 20 --hidden_dim 128 --noise_dim 100

fixed generation

random generation

data directory tree

.
└── data
    └── CelebA
        └── img_align_celeba
            ├── 000001.jpg
            ├── 000002.jpg
            ├── ...
            └── 202599.jpg
    ├── custom_dataset
        └── folder1
            ├── image1.jpg
            ├── ...
    └── ...

References

1. EBGAN : Energy-based Generative Adversarial Network(arxiv:1609.03126)