PyTorch implementation of the paper: Iterative Feedback Network for Unsupervised Point Cloud Registration.

Our model is trained with the following environment:

• Ubuntu 20.04
• Python 3.8
• PyTorch 1.8.1 with torchvision 0.9.1 (Cuda 11.1) Other required packages can be found in requirements.txt.

Clone the repository and build the ops:

git clone https://github.com/IvanXie416/IFNet.git
cd IFNet
cd pointnet2 && python setup.py install && cd ../

The datasets can be downloaded from ModelNet40, 7Scenes, ICL-NUIM and KITTI.

The pre-trained models can be downloaded from Google Drive.

To train a model:

• Modify the 'gaussian_noise', 'unseen', 'data_file', 'dataset_path', 'root' specified in folder 'config' and then do training:

  CUDA_VISIBLE_DEVICES=0 python main.py ./config/train.yaml
  CUDA_VISIBLE_DEVICES=0 python main.py ./config/train7.yaml
  CUDA_VISIBLE_DEVICES=0 python main.py ./config/train-icl.yaml
  CUDA_VISIBLE_DEVICES=0 python main.py ./config/train-k.yaml
  

To test a model:

• Please download the pre-trained models, modify the 'model_path', 'eval' in folder 'config' and then do testing:

  CUDA_VISIBLE_DEVICES=0 python main.py ./config/train.yaml
  CUDA_VISIBLE_DEVICES=0 python main.py ./config/train7.yaml
  CUDA_VISIBLE_DEVICES=0 python main.py ./config/train-icl.yaml
  CUDA_VISIBLE_DEVICES=0 python main.py ./config/train-k.yaml
  

If you find our work useful in your research, please consider citing:

@article{xie2024iterative,
  title={Iterative Feedback Network for Unsupervised Point Cloud Registration},
  author={Xie, Yifan and Wang, Boyu and Li, Shiqi and Zhu, Jihua},
  journal={IEEE Robotics and Automation Letters},
  year={2024},
  publisher={IEEE}
}

This code is developed heavily relying on RIENet, HRegNet and GMFN. Thanks for these great projects.