13,556 LiDAR scans and 6,235 synchronized RGB images captured in unstructured off-road environments. The dataset provides point-level and pixel-level semantic annotations across 20 terrain classes such as grass, mud, and water.
Use Cases
- Train semantic segmentation models for terrain classification using the 20-class pixel-level labels
- Develop point cloud segmentation architectures using the annotated LiDAR scans
- Implement multi-modal sensor fusion for obstacle avoidance by aligning the RGB camera and LiDAR data
- Evaluate SLAM algorithms using the provided IMU and GPS telemetry
Strengths
- 13,556 LiDAR frames captured with an Ouster OS1-64 sensor
- 6,235 RGB images at 1920x1200 resolution with corresponding semantic masks
- 20 semantic classes including 'puddle', 'mud', 'rubble', and 'barrier'
- Includes high-frequency IMU and GPS data for trajectory estimation