A PyTorch-based toolbox for LiDAR-based 3D object detection, supporting multiple state-of-the-art models and datasets.
OpenPCDet is an open-source PyTorch toolbox for LiDAR-based 3D object detection. It provides a unified codebase for implementing and evaluating state-of-the-art 3D detection models, addressing the need for a standardized framework in autonomous driving and robotics research. The project supports multiple datasets and models, enabling researchers to experiment with various architectures efficiently.
Researchers and engineers working on autonomous vehicles, robotics, and computer vision who need a flexible and comprehensive toolkit for 3D object detection from point clouds.
Developers choose OpenPCDet for its extensive model zoo, clear design pattern, and active community support. It offers a self-contained, well-documented codebase that simplifies the implementation of complex 3D detection algorithms and accelerates experimental workflows.
OpenPCDet Toolbox for LiDAR-based 3D Object Detection.
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
Implements over 10 state-of-the-art 3D detection models like PointRCNN, PV-RCNN, and DSVT, with pretrained weights available for datasets such as KITTI, Waymo, and NuScenes, as detailed in the model zoo tables.
Separates data processing from model architecture using a consistent 3D box definition (x,y,z,dx,dy,dz,heading), making it easy to extend to custom models and datasets, as illustrated in the design pattern diagram.
Supports popular autonomous driving datasets including KITTI, Waymo, NuScenes, ONCE, and Argoverse2, with configuration files and evaluation metrics tailored for each, facilitating cross-dataset research.
Regularly updated with new features, such as the 2023 additions of DSVT for real-time inference and BEVFusion for multi-modal fusion, indicating ongoing community support and cutting-edge integration.
Requires managing specific versions of PyTorch and spconv, with installation steps that can be error-prone for different CUDA environments, as highlighted in the INSTALL.md documentation.
Preparing datasets like Waymo involves multiple steps and large storage requirements, and updates like v0.5.0 necessitated re-preparing data, adding significant setup time and potential compatibility issues.
Some models, such as CaDDN at 774MB, have substantial memory footprints, and training times can exceed 15 hours on standard GPUs, limiting accessibility for teams with limited computational resources.