Sparse PointPillars

This is the official repo for our implementation of Sparse PointPillars: Maintaining and Exploiting Input Sparsity to Improve Runtime on Embedded Systems, accepted to IROS 2022.

It is based on the Open3D-ML codebase.

Datasets Used

For our experiments we used KITTI, a standard 3D object detection datset, and Matterport-Chair, a 3D chair detection task dataset generated from multiple houses in the Matterport3D dataset. Model weights and data files are available on our project page.

Citation

Please cite our work (pdf) if you use Sparse PointPillars.

@article{Vedder2022,
    author    = {Kyle Vedder and Eric Eaton},
    title     = {{Sparse PointPillars: Maintaining and Exploiting Input Sparsity to Improve Runtime on Embedded Systems}},
    journal   = {International Conference on Intelligent Robots and Systems (IROS)},
    year      = {2022},
}