A Python devkit for working with the Boreas and Boreas Road Trip all-weather autonomous driving datasets.
pyboreas is a Python development kit for the Boreas and Boreas Road Trip autonomous driving datasets. It provides tools to load, access, and process multi-sensor data including lidar, camera, radar, and GNSS/IMU data collected in all-weather and challenging road conditions. The kit supports benchmarking tasks like odometry, localization, and 3D object detection for autonomous vehicle research.
Researchers and developers working on autonomous driving, robotics, computer vision, and sensor fusion who need access to high-quality, multi-sensor datasets for algorithm development and evaluation.
It offers a standardized and efficient interface to two comprehensive autonomous driving datasets with accurate timestamps and poses, supporting both sequential and random data access patterns. The inclusion of all-weather and challenging road data makes it valuable for robust perception and localization research.
Devkit for the Boreas autonomous driving dataset.
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
Includes high-end sensors like 128-beam Velodyne lidar, 5MP camera, and Navtech radar, providing comprehensive data for multi-modal perception and fusion research.
Covers all-weather conditions in Boreas and challenging roads in Boreas-RT, enabling robustness testing for autonomous systems in varied environments.
Supports live leaderboards for odometry, localization, and 3D object detection, facilitating direct comparison with state-of-the-art methods.
Each sensor frame has precise timestamps and ground truth poses in ENU coordinates, essential for handling asynchronous measurements and accurate localization.
Downloading requires AWS CLI setup and large storage space, with no lightweight or alternative download options, making initial access barrier-heavy.
Users must explicitly call unload_data() on frames to avoid memory bloat, adding complexity and potential for errors in data processing pipelines.
Test sequences lack ground truth poses, limiting evaluation for tasks like odometry without additional workarounds or assumptions.
Sensors are not synchronous (unlike datasets like KITTI), requiring extra effort for data alignment, which can be non-trivial for beginners.