miccai2022-roigan

Code to train region-guided CycleGANs [1].

• main.py contains training code.
• src/models.py defines GAN generator and discriminators.
• src/utils.py defines utility functions for training and graphing.
• config/ defines .yml configuration files to set experiment parameters.

Installation

The Anaconda environment is specified in environment.yml. The environment can be recreated using,

conda env create -f environment.yml

Tested with single NVIDIA V100 GPU, running Cuda 10.0.130, and PyTorch 1.9.0 with torchvision 0.10.0.

Usage

main.py is the training code, which requires two parameters

• job_number specifies a unique identifier for writing outputs
• config specifies configuration file path

See slurm_submit.sh for example.

Config files

See config/README.md for a description of configuration options.

Data

Experiments performed on CAMELYON16 and data from the Gustave Roussy Institute.

See data/README.md for library building instructions.

Reference

[1] Region-guided CycleGANs for Stain Transfer in Whole Slide Images, Joseph Boyd, Irène Villar, Marie-Christine Mathieu, Eric Deutsch, Nikos Paragios, Maria Vakalopoulou, and Stergios Christodoulidis, MICCAI 2022 (in press) [PDF]

Model outputs