AI City Track 3 Viewer

AI City Challenge 2026 Track 3 — Anomalous Events in Transportation. 44,040 chain-of-thought annotations over 3,670 surveillance videos, drawn from 8 source datasets (SO-TAD, TAD-Benchmark, UCF Crime, HTV, TAD, Vad-R1, Barbados, Accident-Bench). Each video has 12 annotated task instances spanning binary verification (bcq), 4-option MCQ, open QA, scene description, video summarization, temporal localization, temporal description, and causal linkage. Annotations follow the tao-vl-reason-v1.0 envelope schema. This viewer ships 512 videos (64 per source) with all 12 task instances each, totalling 6,144 annotation samples — re-encoded to 480p for in-browser playback. Aggregate stats across the full release are shown at the bottom. Official challenge page: aicitychallenge.org/2026-track3. Annotations on Hugging Face: nvidia/PhysicalAI-Traffic-Anomaly-Reasoning.