IIRP-Net

This is the official pytorch implementation of "IIRP-Net: Iterative Inference Residual Pyramid Network for Enhanced Image Registration" (CVPR 2024), written by Tai Ma, Suwei Zhang, Jiafeng Li and Ying Wen. Paper link: https://openaccess.thecvf.com/content/CVPR2024/html/Ma_IIRP-Net_Iterative_Inference_Residual_Pyramid_Network_for_Enhanced_Image_Registration_CVPR_2024_paper.html

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{Ma_2024_CVPR,
    author    = {Ma, Tai and Zhang, Suwei and Li, Jiafeng and Wen, Ying},
    title     = {IIRP-Net: Iterative Inference Residual Pyramid Network for Enhanced Image Registration},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2024},
    pages     = {11546-11555}
}

The overall framework of the code and the Swin Transformer module are based on VoxelMorph , 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: For RP-Net

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

The test results are:

For IIRP-Net