cs-xiao / targan

TarGAN - Official PyTorch Implementation (MICCAI 2021)

TarGAN

PyTorch implementation of the MICCAI 2021 paper "TarGAN: Target-Aware Generative Adversarial Networks for Multi-modality Medical Image Translation". (It will be updated soon~)

Dependencies

• Python 3.7+
• PyTorch 1.6.0+
• Visualdl 2.0.5+ (optional for log)

Downloading datasets

To download the CHAOS dataset from the CHAOS challenge website. Then, you need to create a folder structure like this:

datasets/chaos2019/
├── train
│   ├── ct
│   │   ├──1
│   │   ├──...(patient index)
│   ├── t1
│   │   ├──3
│   │   ├──...(patient index)
│   ├── t2
│   │   ├──5
│   │   ├──...(patient index)
├── test
│   ├── ct
│   │   ├──2
│   │   ├──...(patient index)
│   ├── t1
│   │   ├──4
│   │   ├──...(patient index)
│   ├── t2
│   │   ├──6
│   │   ├──...(patient index)

Training networks

To train TarGAN on CHAOS, run the training script below.

# Train TarGAN using the CHAOS dataset
python train.py -datasets chaos -save_path yours -epoch 50\
               -c_dim 3 -batch_size 4 -lr 1e-4 -ttur 3e-4\
               -random_seed 666