SSCBench: Large-Scale 3D Semantic Scene Completion for Autonomous Driving

SSCBench is a large-scale benchmark for 3D semantic scene completion in autonomous driving scenarios, presented at IROS 2024 by the ai4ce research group. The dataset provides 3D occupancy grid maps and semantic labels to facilitate 3D scene understanding from 2D or 3D sensor inputs. It was released in 2024 to address the gap in standardized evaluation for semantic scene completion tasks.

Use Cases

• Training 2D-to-3D semantic lifting models using occupancy grid maps
• Evaluating 3D scene understanding algorithms for autonomous vehicle navigation
• Benchmarking semantic scene completion accuracy across different sensor modalities

Strengths

• Peer-reviewed benchmark from IROS 2024
• Specialized for 3D Semantic Scene Completion (SSC) tasks
• Includes occupancy-grid-map labels for complex scene understanding

Limitations

• Specific record counts and dataset size are not explicitly detailed in the primary metadata
• High computational overhead typical of 3D voxel-based processing

Provenance

Source

ai4ce (New York University)

Collection Method

Sensor-based data collection with expert-level semantic annotation for 3D volumes

Freshness

Updated as of April 2025; represents state-of-the-art research from the IROS 2024 conference.