Skip to content

Latest commit

 

History

History
30 lines (28 loc) · 1.86 KB

README.md

File metadata and controls

30 lines (28 loc) · 1.86 KB
Raw

PIViT

This is the official pytorch implementation of "PIViT: Large Deformation Image Registration with Pyramid-Iterative Vision Transformer" (MICCAI 2023), written by Tai Ma, Xinru Dai, Suwei Zhang and Ying Wen. Paper link: https://link.springer.com/chapter/10.1007/978-3-031-43999-5_57 image

Environment

We reimplemented the code on pytorch 1.13 and python 3.7.15.

Dataset

We performed retraining，validation and testing on the Mindboggle dataset.

We provide pre-trained models on the Mindboggle dataset, trained with two subsets, NKI-RS and NKI-TRT, with images cropped to the size of (160, 192, 160).

Citation

If you use the code in your research, please cite:

@inproceedings{ma2023pivit,
  title={PIViT: Large Deformation Image Registration with Pyramid-Iterative Vision Transformer},
  author={Ma, Tai and Dai, Xinru and Zhang, Suwei and Wen, Ying},
  booktitle={International Conference on Medical Image Computing and Computer-Assisted Intervention},
  pages={602--612},
  year={2023},
  organization={Springer}
}

The overall framework of the code and the Swin Transformer module are based on VoxelMorph and TransMorph, whose contributions are greatly appreciated.

Test

We provide the pre-trained model and two images for testing from the MMRR subset of the Mindboggle dataset. You can test it with the following code:

python test.py --scansdir   data/vol --labelsdir  data/seg --dataset mind --labels  data/label_mind.npz --model 0980.pt --gpu 0

The test results are: image