we investigate unstructured scene understanding through 3D semantic occupancy prediction, which is used to detect irregular obstacles in unstructured scenes, and road surface elevation reconstruction, which characterizes the bumpy and uneven conditions of road surfaces. The dataset provides detailed annotations for 3D semantic occupancy prediction and road surface elevation reconstruction, offering a comprehensive representation of unstructured scenes. In addition, trajectory and speed planning information is provided to explore the relationship between perception and planning in unstructured scenes. Natural language descriptions of scenes are also provided to explore the interpretability of autonomous driving decision-making.
- [2025/5/10] UnScenes3D Dataset v1.0 Released
Dataset construction framework and future outlook: (a) Data processing. (b) Data label visualization. (c) Scene text description. (d) Future work outlook.
Please refer to PIPLINE for more details.
please download unscenes3d-mini（14 scenes） from Releases, and put it in a folder named data,with the following structure:
./data/unscenes3d/
├── calibs/ # Calibration information for sensors
├── images/ # Synchronized frame image data
│ ├── 1689903584.278848.jpg
│ └── ...
├── clouds/ # HAP-synchronized frame point cloud data
│ ├── 1689903584.278848.bin
│ └── ...
├── occ/ # 3D semantic occupancy prediction labels
├── elevation/ # Road elevation labels
├── depths/ # Depth estimation labels
├── imagesets/ # Dataset splits for training, validation, and testing
│ ├── scene_info.json
├── localmap_clouds/ # Dense point cloud map of local environment
├── vehicle_infos/ # Ego vehicle's pose, speed, and acceleration information
└── image_caption/ # Language-based scene descriptions
Many thanks to these excellent open source projects:
