MMNet

A PyTorch implementation of "MMNet: A medical image-to-image translation network based on manifold value correction and manifold matching".

---

Environment

1. Linux(ubuntu)
2. python=3.7.3
3. torch=1.7.1+cu92
4. tqdm=4.32.1
5. opencv-python=4.5.3.56
6. pyyaml=5.1.1
7. visdom=0.1.8.9

---

Demo

1. The pre-trained models are in "checkpoints/MMNet";

2. Download patial test samples from GoogleDrive, then put them into corresponding dir ("datasets/BraTs2015/val or datasets/OASIS3/val");

3. Modify the MMNet.yaml in "Yaml/MMNet.yaml";

   Config for test.

   run_name: 'MMNet/BraTs2015/'
   dataset: BraTs2015
   val_dataroot: 'datasets/BraTs2015/val'
   input_nc: 1

   run_name: 'MMNet/OASIS3/'
   dataset: OASIS3
   val_dataroot: 'datasets/OASIS3/val'
   input_nc: 3

4. python test.py

---

Acknowledgments

Code borrows from Reg-GAN and pytorch-manifold-matching. The distribution generator and distribution corrector is modified from Reg-GAN and VoxelMorph.