Benchmark RadarSP-TransientBench
SP-TransientBench (STB) is a real-captured multi-task benchmark for single-photon perception. It comprises 10 diverse scenes and 10,297 views captured with a solid-state single-photon LiDAR at 256×192 resolution, providing full time-of-flight histograms with multi-return behavior, calibrated camera poses, and standardized metadata. The benchmark evaluates depth estimation, multi-view reconstruction, and 3D semantic understanding, with 13-class semantic annotations for selected scenes. Dedicated data splits and evaluation protocols are provided for each task.
- Released
- 2026-06-16
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
Existing single-photon perception studies rely on simulated or small-scale captures, lacking a systematic real-world evaluation. STB provides a consistent and reproducible benchmark across multiple 3D vision tasks, enabling comparative assessment of algorithms in photon-starved scenarios and advancing active 3D perception research.
Motivation
Single-photon LiDAR (SPL) based on single-photon avalanche diode (SPAD) sensing enables time-resolved photon measurements with extreme sensitivity, offering unique potential for active 3D perception in photon-starved scenarios.However, real-world single photon perception remains fundamentally challenging due to unique measurement noise and complex multi-return transient phenomena, which jointly complicate geometric…
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.