guochengqian/openpoints

OpenPoints: a library for easily reproducing point-based methods for point cloud understanding. The engine for [ASSANet](https://arxiv.org/abs/2110.10538), [PointNeXt](https://arxiv.org/abs/2206.04670)

What it solves

OpenPoints is designed to provide a standardized library for benchmarking and reproducing point-based methods for point cloud understanding. It eliminates the need to rewrite complex point cloud operations from scratch, allowing researchers to compare different models fairly and reproduce results more easily.

How it works

The library acts as an engine that provides a wide range of basic operations—such as graph convolutions, self-attention, farthest point sampling, and ball query—which are used to build various point cloud networks. It uses a configuration-based system (YAML files) to easily build models, optimizers, schedulers, and loss functions without changing the code.

Who it’s for

It is primarily for researchers and developers working on 3D point cloud understanding and computer vision, specifically those using the PyTorch ecosystem.

Highlights

  • Broad Model Support: Supports reproducing several representative networks including PointNet, DGCNN, DeepGCN, PointNet++, ASSANet, PointMLP, PointNeXt, Pix4Point, and PointVector.
  • Configuration-Driven: Allows users to train and validate different models on various tasks by simply modifying YAML configuration files.
  • Extensible Framework: Provides a comprehensive set of basic operations that make it easy to build new networks.
  • CUDA-Accelerated: Includes specialized C++/CUDA operators for high-performance point cloud processing.

Related

  • Project
    isl-org/Open3D-ML

    An extension of Open3D that provides tools, pretrained models, and pipelines for 3D machine learning tasks like semantic segmentation and object detection.

  • Project
    openml/automlbenchmark

    A framework for evaluating and comparing open-source AutoML systems using standardized environments and curated datasets to ensure reproducible research.

  • Project
    isl-org/Open3D

    Open3D is a modern library for 3D data processing that provides optimized tools for point cloud and mesh manipulation, visualization, and 3D machine learning.

  • Project
    PointCloudLibrary/pcl

    A large-scale open-source library for 2D/3D image and point cloud processing, used widely in robotics and 3D perception.

  • Project
    benchopt/benchopt

    A benchmarking suite for machine learning and optimization workflows designed to ensure reproducibility and comparability across solvers written in multiple programming languages.