40,000 voxelized frames and 18 semantic categories define this dataset for the CVPR 2023 3D Occupancy Prediction Challenge. It provides dense 3D annotations for outdoor driving environments to support the development of spatial perception algorithms.
Use Cases
- Train 3D occupancy networks to predict voxel labels using the semantics and occupancy features
- Benchmark multi-view 3D perception models against the ground-truth voxel grid
- Develop sensor-fusion techniques for dense 3D scene reconstruction
Strengths
- Includes 3D voxel grids with occupancy and semantic labels for 18 distinct classes
- Features multi-view camera data aligned with 3D spatial annotations
- Provides temporal sequences of 3D occupancy frames for dynamic scene understanding