Research map / 3D vision / Depth and stereo
SP-TransientBench: A Real-Captured Single Photon Perception Benchmark
Hongzhou Dong, Zili Zhang, Ziting Wen, Yiheng Qiang, Runrong Deng, Wenle Dong, Ziwen Jiang, Xinyang Li, Rui Lu, Shuoyao Sun, Wenyu Wang, Ziyi Xia and 3 more
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add histogram preprocessing before point-cloud segmentation (ablation) | PointNet++ · STB 3D semantic segmentation | mIoU: 49.16 -> 51.31 | Improved |
| add histogram preprocessing (thresholding) before segmentation (ablation) | PointMLP · STB 3D semantic segmentation | mIoU: 49.11 -> 52.22 | Improved |
| add SSPINET preprocessing before segmentation (ablation) | PointNeXt · STB 3D semantic segmentation | mIoU: 41.05 -> 41.54 | No clear effect |
| add Matched Filtering preprocessing (ablation) | PointNeXt · STB 3D semantic segmentation | OA: 88.45 -> 87.07 | Hurt |
| swap PointNet++ backbone for PointNeXt with no preprocessing (ablation) | segmentation pipeline · STB 3D semantic segmentation | mIoU: 49.16 -> 41.05 | Hurt |
| pretrain on simulated ScanNet transients then finetune, 10% real data (ablation) | PointNet++ · STB segmentation, 10% real data | mIoU: 33.19 -> 36.57 | Improved |
| pretrain on simulated ScanNet transients then finetune, 20% real data (ablation) | PointNet++ · STB segmentation, 20% real data | mIoU: 37.98 -> 41.39 | Improved |
| pretrain on simulated ScanNet transients then finetune, 100% real data (ablation) | PointNet++ · STB segmentation, 100% real data | mIoU: 52.24 -> 51.20 | No clear effect |
| use learning-based SSPINET instead of model-based Shin for depth | depth estimation · STB depth estimation | Recall 1 bin (%): 25.97 -> 44.27 | Improved |
| loosen recall tolerance from 1 bin to 5 bins | SSPINET · STB depth estimation | Recall (%): 44.27 -> 93.68 | Improved |
| TransientNeRF versus Transientangelo | multi-view reconstruction · STB multi-view reconstruction (10 views) | Histogram PSNR: 45.4251 vs 41.5735 | Mixed |
Closest papers by meaning
- Reconstructing Transient Images From Single-Photon Sensors (CVPR 2017 · similarity 0.89)
- PlatoNeRF: 3D Reconstruction in Plato's Cave via Single-View Two-Bounce Lidar (CVPR 2024 · similarity 0.88)
- Eulerian Single-Photon Vision (ICCV 2023 · similarity 0.88)
- Opportunistic Single-Photon Time of Flight (CVPR 2025 · similarity 0.87)
- Separating the Wheat from the Chaff: Spatio-Temporal Transformer with View-interweaved Attention for Photon-Efficient Depth Sensing (AAAI 2025 · similarity 0.86)
- Towards 3D Vision with Low-Cost Single-Photon Cameras (CVPR 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
