Abstract:Grounding traffic accidents in real CCTV footage is a rare-event problem where training on labeled accident video is often prohibited, yet accurate joint localization in time, space, and collision type is required. We present a no-fine-tuning pipeline that elicits this joint output from frozen vision-language models through two ideas. First, a coarse-to-fine two-pass decomposition: a full-video pass at 1 fps produces a coarse (t, x, y, c) tuple, then a second pass at 5 fps within a +/- 3 s window refines time and location, with two deterministic confidence gates that revert to the coarse estimate on boundary hedges or edge-clamped coordinates. Second, a specialist role assignment: Qwen3-VL-Plus handles grounding, Gemini 3.1 Flash-Lite handles typing on a centered video clip. On the ACCIDENT@CVPR 2026 benchmark (2,027 real CCTV videos) we reach ACC^S = 0.539 (95% CI [0.525, 0.553]): +0.127 over the benchmark paper's best-of-baselines oracle (0.412), +0.143 over the strongest single-VLM baseline (Molmo-7B, 0.396), and +0.250 over the naive baseline (0.289). The VLM path uses up to three API calls per video (17% fall back to physics on API failures); the full run costs ~$20.
| Comments: | Accepted at CVPR 2026 AUTOPILOT Workshop (Non-Archival Track). 7 pages (4 main + references + appendix), 3 figures, 5 tables |
| Subjects: | Computer Vision and Pattern Recognition (cs.CV) |
| Cite as: | arXiv:2605.01512 [cs.CV] |
| (or arXiv:2605.01512v2 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2605.01512 arXiv-issued DOI via DataCite |
From: Jiantang Huang [view email]
[v1]
Sat, 2 May 2026 16:05:44 UTC (377 KB)
[v2]
Sat, 23 May 2026 18:49:50 UTC (378 KB)
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。