- Data: Dense surface normals for CAM_BACK_RIGHT.
- Resolution: 448x800.
- Field of View: Rear-right diagonal perspective.
- Use Case: Completes the 360-degree environmental geometric reconstruction.
- Context: Final piece of the 6-camera dense normal sensor suite for nuScenes.
- Data: Dense surface normals for CAM_BACK_LEFT.
- Resolution: 448x800.
- Field of View: Rear-left diagonal perspective.
- Use Case: Essential for blind-spot geometry analysis and full-vehicle surround reconstruction.
- Part of: 360-degree dense normal suite for nuScenes.
- Data: Dense surface normals for CAM_BACK.
- Resolution: 448x800.
- Field of View: Direct rear-facing perspective (180° degree opposite to CAM_FRONT).
- Use Case: Crucial for rear-end obstacle detection and full-scene 3D surface reconstruction.
- Progress: Continuing the 360-degree dense normal suite upload.
- Data: Dense surface normals for CAM_FRONT_RIGHT.
- Resolution: 448x800.
- Field of View: Right-forward facing (~55° FOV).
- Use Case: Complements CAM_FRONT for wider forward-facing geometric perception.
- Progress: Part of the 360-degree dense normal suite.
- Data: Dense surface normals for CAM_FRONT.
- Resolution: 448x800.
- Source: Processed from nuScenes primary camera sensor.
- Use Case: Geometric perception, surface orientation, and 3D reconstruction.
- Added CAM_FRONT_RIGHT semantic masks in .tar.gz format. - Processing Pipeline: GroundedSAM static masks + SegFormer dynamic objects. - Encoding Specification: * 1-16: NuScenes standard static semantic classes. * 101-116: Dynamic object classes (Semantic ID + 100). * 255: Sky mask (preserved for downstream high-fidelity processing). - Data alignment: Verified with nuScenes CAM_FRONT_RIGHT sensor samples. - Progress: 6/6 cameras completed. Full 360-degree coverage for the nuScenes dataset is now available.
- Added CAM_BACK semantic masks compressed in .tar.gz format. - Processing Pipeline: GroundedSAM (static) merged with SegFormer (dynamic). - Encoding Scheme: * 0: Ignore/Background * 1-16: NuScenes standard static classes * 101-116: Dynamic classes (Semantic ID + 100) * 255: Sky mask (preserved for downstream extraction) - Sensor Context: Part of the 360-degree camera suite in nuScenes. - Data integrity verified against original CAM_BACK JPG samples.
- Add pre-processed sparse lidar depth data for all 6 camera views. - Format: Compressed numpy archives (.npz) containing depth and validity masks. - Resolution: 900x1600 (consistent with nuScenes camera sensors). - Source: Projected from LIDAR_TOP to camera planes using nuScenes-devkit. - Purpose: Support downstream tasks like Depth Completion or 2D-3D Grounding.
