ChunyuanLI / DiscoGAN

Tensorflow implementation of "Learning to Discover Cross-Domain Relations with Generative Adversarial Networks"

DiscoGAN

Tensorflow implementation of Learning to Discover Cross-Domain Relations with Generative Adversarial Networks.

Recent development

Our NIPS paper ALICE improves DiscoGAN/CycleGAN/DualGAN, please see its code repo, and comparison on cartoon generation.

Alice4Alice: ALICE algorithms for painting the cartoon of Alice's Adventures in Wonderland

Prerequisites

• Python 3.5
• Tensorflow 1.0
• Others

5-GMM to 2-GMM

The demo is tested a toy dataset, with domain X as 5-component GMM, and domain Z as 2-component GMM. It can be easily extended to other GMM settings, and real dataset.

To train:

$ python DiscoGAN_main.py

The reuslts:

Links

• Official PyTorch implementation (https://github.com/SKTBrain/DiscoGAN)
• Taehoon Kim's PyTorch implementation (https://github.com/carpedm20/DiscoGAN-pytorch)