发表机构
NVIDIA; Santa Clara University（英伟达; 圣克拉拉大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
该研究提出TAR与TAR-Bench数据集，用于训练评估超越异常检测的视频-语言模型，经实验验证其多任务微调可提升模型推理能力，且为2026年AI城市挑战赛第3赛道提供官方数据。
AI 中文摘要
我们提出了TAR（Traffic Anomaly Reasoning，交通异常推理）和TAR-Bench数据集，这是用于训练和评估视频-语言模型（超越异常检测）的资源。TAR包含来自8个公开数据集的3670个CCTV视频（约26小时）的10个任务下的44040条思维链训练标注。其评估组件TAR-Bench包含960条人工策划的测试标注，对应从17个公开YouTube视频中剪辑的80个保留片段。TAR的训练标注由MAVEN生成，MAVEN在生成问答对和推理轨迹前，将多尺度视频证据整合为结构化事件描述。在TAR-Bench上，11个视觉-语言模型的结果表明，出色的问答准确率无法可靠预测时间或场景推理能力。在TAR上进行多任务微调可带来稳定提升，完整的10任务模型的综合得分比其零样本基线提高了21.4个百分点。TAR和TAR-Bench是2026年AI City Challenge（AI城市挑战赛）第3赛道的官方训练和域内评估数据。该数据集可通过此URL获取。
英文摘要
Detecting a traffic anomaly does not establish whether a video-language model can explain what happened, localize it in time, or identify its causes. We introduce TAR (Traffic Anomaly Reasoning) and TAR-Bench, paired resources for training and evaluating these complementary capabilities across 10 tasks spanning question answering, temporal reasoning, and scene understanding. TAR contains 44,040 automatically generated annotations with chain-of-thought traces for 3,670 CCTV videos from eight public datasets. TAR-Bench provides 960 human-curated annotations for 80 held-out clips from 17 public YouTube videos. Evaluation of eleven vision-language models reveals a gap between question-answering performance and temporal or scene reasoning. Progressively adding task groups during supervised fine-tuning improves aggregate performance on both Cosmos-Reason2-8B and Qwen3-VL-8B-Instruct. Training on all 10 tasks raises their mean benchmark scores from 34.3 to 55.7 and from 30.9 to 53.9, respectively. These results support joint training across complementary tasks as a promising approach to traffic anomaly understanding, while highlighting persistent limitations in temporal precision and causal attribution. TAR and TAR-Bench serve as the official training and in-domain evaluation resources for AI City Challenge 2026 Track 3.The dataset is available at https://huggingface.co/datasets/nvidia/PhysicalAI-Traffic-Anomaly-Reasoning
