Synthetic Medical Images from Dual Generative Adversarial Networks

Code is split into two stages: a segmentation-mask-generating DCGAN, and an image-to-image translator using pix2pix.

Paper: https://arxiv.org/abs/1709.01872

SynthMed (Repository for GAN-produced synthetic medical images): https://synthmed.github.io/

Pipeline

Prerequisites

• Python 2 and 3
• numpy
• TensorFlow 1.0+
• Keras
• Preprocessed dataset

Acknowledgements

Stage-I GAN based on: https://github.com/carpedm20/DCGAN-tensorflow

Stage-II GAN based on: https://github.com/ray0809/pix2pix

Authors: John Guibas, Tejpal Virdi, Peter Li