A collection of high-performance GICP-based point cloud registration algorithms with multi-threaded and GPU-accelerated implementations.
fast_gicp is a collection of high-performance point cloud registration algorithms based on the Generalized Iterative Closest Point (GICP) method. It provides multi-threaded and GPU-accelerated implementations that significantly outperform standard PCL implementations, enabling real-time 3D mapping and localization.
Robotics researchers, autonomous vehicle developers, and computer vision engineers working with LiDAR data who need fast and accurate point cloud alignment for SLAM, 3D reconstruction, or object detection applications.
Developers choose fast_gicp because it provides drop-in replacements for PCL's GICP that are 2-10x faster through parallelization and GPU acceleration, while maintaining the same interfaces and accuracy for seamless integration into existing pipelines.
A collection of GICP-based fast point cloud registration algorithms
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 multi-threaded GICP achieving ~40 FPS and VGICP at ~70 FPS, significantly outperforming standard PCL implementations as shown in benchmarks.
Offers CUDA-based VGICP at ~120 FPS and NDT at ~500 FPS, enabling real-time performance for demanding applications like autonomous vehicle mapping.
Provides identical interfaces to PCL's GICP, allowing seamless integration as a replacement in existing point cloud processing workflows without code changes.
Includes full Python bindings with both functional (pygicp.align_points) and class-based (pygicp.FastGICP) interfaces, facilitating rapid prototyping and scripting.
Requires PCL, Eigen, OpenMP, Sophus, nvbio, and optional CUDA, with installation notes highlighting challenges on macOS and specific Ubuntu versions, increasing setup overhead.
The README starts by promoting small_gicp as twice as fast with minimum dependencies, indicating fast_gicp may be less optimal for new projects or receive reduced maintenance.
Tested primarily on Ubuntu 18.04/20.04 with CUDA 11.1, which can cause portability issues and require additional tuning for other environments or hardware setups.