KPConvX: Modernizing Kernel Point Convolution with Kernel Attention

[Blog]      [ArXiv]

This repo is the official project repository of the paper KPConvX: Modernizing Kernel Point Convolution with Kernel Attention.

Highlights

• Nov 14, 2024: We release the trained models for ScanObjectNN and S3DIS datasets
• Nov 14, 2024: KPConvX repository is officially realeased. We provide two implementation for KPConvX, an standalone training pipeline, and wrappers to use it in the Pointcept library.
• Feb 28, 2024: KPConvX is accepted by CVPR 2024! 🎉

Setup

The setup instructions are different if you plan to use the Standalone or the Pointcept version:

• Setup the standalone version.
• Setup the Pointcept version.

Citation

If you found this code useful, please cite the following paper:

@inproceedings{thomas2024kpconvx,
  title={KPConvX: Modernizing Kernel Point Convolution with Kernel Attention},
  author={Thomas, Hugues and Tsai, Yao-Hung Hubert and Barfoot, Timothy D and Zhang, Jian},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={5525--5535},
  year={2024}
}

Acknowledgements

Our codebase is built using multiple opensource contributions, please see ACKNOWLEDGEMENTS for more details.