From Detection to Understanding: TAR and TAR-Bench for Multi-Task Traffic Anomaly Reasoning

David C. Anastasiu; Han Zhang; Tomasz Kornuta; Varun Praveen; Vidya N. Murali; Yilin Zhao; Zaid Pervaiz Bhat; Zheng Tang

REVIEW 2 major objections 4 minor 34 references

From Detection to Understanding: TAR and TAR-Bench for Multi-Task Traffic Anomaly Reasoning

T0 review · 2 major / 4 minor · reviewed 2026-08-14 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper claims that traffic-anomaly understanding requires multi-task chain-of-thought supervision: question-answering accuracy does not predict temporal or scene reasoning, and ten-task fine-tuning raises aggregate benchmark scores by…

desk verdict A transparent, useful benchmark whose multi-task fine-tuning gains are real on closed-form tasks, but the open-ended reasoning gains need an independence check before the headline claim is fully trusted. read the letter →

arxiv 2608.10317 v1 pith:TQD6MYEY submitted 2026-08-10 cs.CV

Han Zhang , Yilin Zhao , Zaid Pervaiz Bhat , Zheng Tang , Varun Praveen , Vidya N. Murali , David C. Anastasiu , Tomasz Kornuta This is my paper · ORCID

classification cs.CV

keywords trafficanomalyreasoningvideodetectionvision-languagemodelschain-of-thoughtmulti-taskfine-tuningtemporalsceneunderstandingquestionanswering

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper claims that teaching models to understand traffic anomalies requires moving beyond detection. It introduces TAR, a training corpus of 44,040 chain-of-thought annotations across ten tasks on 3,670 surveillance videos, and TAR-Bench, a human-curated benchmark of 960 annotations on 80 held-out clips. The central empirical finding is that strong question-answering accuracy does not reliably predict temporal or scene reasoning, and that multi-task fine-tuning on TAR raises the aggregate benchmark score by 21.4 points over the zero-shot baseline, with the largest gains on the reasoning tasks that detection-style supervision ignores. If the claim holds, standard anomaly-detection accuracy is measuring the wrong capability for real-world monitoring.

What carries the argument

The load-bearing mechanism is MAVEN, a three-stage annotation pipeline that first collects multi-scale video evidence (a global scene caption, timestamped event captions, and short-chunk captions), consolidates that evidence into a structured spatio-temporal event description, and then uses that description as the sole context to generate question-answer pairs with explicit reasoning traces for ten tasks. The ten tasks are organized into three groups—question answering, temporal reasoning, and scene understanding—so training forces the model to connect event recognition to temporal, causal, and scene-level evidence. Fine-tuning uses dual supervision: each example is seen once with loss on the answer only and once with loss on the full reasoning trace and answer, teaching the model both to answer directly and to produce structured explanations. This shared taxonomy, with a reasoning trace for every annotation, is what makes the multi-task fine-tuning claim testable.

What would settle it

Draw a random sample of model outputs on the temporal-localization, causal-linkage, and scene-description tasks, have human annotators judge logical and factual correctness without seeing the benchmark references, and compare that ranking with the paper's text-similarity ranking; if the rankings diverge substantially, the reported gains do not establish genuine reasoning improvement.

Watch

Extended reading notes

Core claim

The paper's central claim is that detection-style accuracy is a poor proxy for video reasoning, and that a deliberately structured multi-task corpus can close the gap. The best model without prior exposure to the training data averages 39.5% on TAR-Bench, while a much larger proprietary model scores below smaller open-weight models, showing that scale alone does not produce reasoning. Fine-tuning an 8-billion-parameter model on the five question-answering tasks raises its average from 34.3% to 49.6%; adding the three temporal-reasoning tasks raises it to 53.5%; adding the two scene-understanding tasks raises it to 55.7%, a total gain of 21.4 points. The same monotonic pattern appears in a second model family, and answer-only supervision scores lower than supervision with reasoning traces. The paper interprets these results as evidence that complementary supervision, not scale, is what moves a model from flagging an anomaly to explaining it.

Load-bearing premise

The results stand on the assumption that the benchmark's human-curated reference answers, scored by a text-similarity metric, measure reasoning quality rather than stylistic resemblance to the labels the annotation pipeline generates.

Editorial extensions

If this is right

Anomaly-detection leaderboards that report only classification accuracy will misrank systems for real-world deployment, so temporal, causal, and scene scores need to be reported separately.
Adding task groups during fine-tuning improves unseen reasoning tasks that were never supervised, indicating that multi-task annotations act as a general video-representation regularizer.
Chain-of-thought inference, which hurts zero-shot performance, becomes comparable to direct answering after fine-tuning on reasoning traces, aligning the model's reasoning format with evaluation.
The monotonic gains reproduce across two open-weight model families, suggesting the multi-task effect is not an artifact of one architecture.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The headline 21-point gain is measured with a text-similarity metric that the paper itself notes measures semantic similarity rather than logical correctness; a human audit of logical correctness could show the true reasoning gain is smaller or larger.
Because all TAR-Bench clips come from only 17 public videos, source-level correlation may inflate scores; grouping evaluation clips by source would test how much of the ranking reflects model ability versus scene diversity.
The same annotation pipeline could be applied to other surveillance domains such as retail or public spaces to test whether the multi-task gain transfers, a transfer the paper explicitly leaves unevaluated.
The finding that question-answering accuracy does not predict reasoning suggests future video-reasoning benchmarks should weight temporal and causal tasks more heavily, or the community will keep optimizing the wrong metric.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

First thing to know: this is a dataset/benchmark paper that does its job honestly. TAR's ten-task taxonomy with explicit chain-of-thought traces, the separation of a large auto-labeled training corpus from a human-curated evaluation set, and the AI City Challenge 2026 Track 3 integration give it real value. The release is public, CC-BY-4.0, with retrieval scripts rather than redistributed video, which is reproducible in the way that matters.

The strongest evidence is the progressive fine-tuning ablation on two backbones. On CR2-8B the mean rises monotonically from 34.3% to 49.6% (QA), 53.5% (QA+TR), and 55.7% (all 10). Qwen3-VL-8B shows the same pattern: 30.0% to 48.6%, 51.1%, 53.9%. Gains on binary QA and MCQ are large and measured by accuracy, so those are solid. The separate finding that QA accuracy is a poor predictor of temporal or scene reasoning is supported by the zero-shot table and is worth having on record.

Soft spots, in proportion. First, TAR-Bench references are MAVEN drafts corrected by four experts, with median string similarity 0.95 to the originals. Fine-tuning on TAR trains on MAVEN-generated outputs, so gains on the seven open-ended tasks scored by BERTScore may partly reflect dialect or style matching rather than improved reasoning. The authors concede that BERTScore measures semantic similarity, not logical correctness, and they flag this in the limitations. I do not think it invalidates the dataset, but it does mean the 'reasoning beyond detection' headline rests on an unverified step. An independent-reference or human-scored check on the open-ended tasks would settle it. Second, TAR-Bench is 80 clips from 17 YouTube videos; source-level correlation is real and acknowledged. Third, the 44K training labels are unaudited VLM outputs. The paper says so plainly, which I respect, but users should not treat them as ground truth.

Minor notes: the VAD-R1 near-constant-prediction analysis is a nice diagnostic, and the temporal-localization near-miss discussion (median 0.68s boundary gap among zero-IoU cases) is a fair argument for adding boundary-distance diagnostics.

Bottom line: this paper should go to serious peer review, not desk rejection. The artifact is useful regardless, and the main empirical pattern is consistent across two backbones. A good referee should ask for the independence check on open-ended scoring, but the core dataset contribution holds up.

Referee Report

2 major / 4 minor

Summary. The paper introduces TAR, a multi-task training corpus containing 44,040 chain-of-thought annotations across 10 tasks for 3,670 CCTV videos drawn from eight public datasets, and TAR-Bench, a human-curated evaluation benchmark of 960 annotations over 80 held-out clips from 17 YouTube videos. Training annotations are generated by the MAVEN pipeline (Gemini 3.1 Pro for evidence extraction and Gemma-4-31B for task generation), while TAR-Bench annotations are MAVEN drafts corrected by four expert reviewers. The paper evaluates eleven vision-language models in zero-shot mode and reports that strong binary/multiple-choice QA accuracy does not reliably predict temporal or scene reasoning performance. Progressive multi-task fine-tuning on TAR yields aggregate score gains of 21.4 points on Cosmos-Reason2-8B and 23.9 points on Qwen3-VL-8B-Instruct, with a cross-backbone replication of the monotonic task-group improvement. The paper also provides a diagnostic error analysis and describes the integration of TAR and TAR-Bench into AI City Challenge 2026 Track 3.

Significance. If the results hold, TAR and TAR-Bench make a useful contribution by moving traffic video benchmarks beyond binary anomaly detection toward structured multi-task video-language reasoning, with a clear task taxonomy, explicit chain-of-thought traces, and a human-corrected evaluation set. The paper is transparent about its annotation pipeline and reports detailed human-correction statistics. The cross-backbone replication of the multi-task fine-tuning gain is a strength, as is the public release of the dataset. The finding that QA accuracy does not predict temporal or scene reasoning is interesting and likely to influence benchmark design. However, the central claim that multi-task chain-of-thought supervision genuinely improves video-language reasoning beyond simple detection rests substantially on BERTScore-based open-ended task gains whose references are MAVEN-drafted; this step needs independent validation before the claim can be taken as established.

major comments (2)

[§4.1, §6, Appendix A] The seven open-ended tasks are scored with BERTScore against TAR-Bench references that were drafted with MAVEN and then human-corrected, with Appendix A reporting a median string similarity of 0.95 between original and corrected answers. Because the TAR training labels are generated by the same MAVEN pipeline, fine-tuning may improve BERTScore by matching the annotation dialect and output distribution of MAVEN rather than by genuinely improving temporal, causal, or scene reasoning. The aggregate fine-tuning gain of 21.4 points in Table 4 therefore mixes objective closed-form gains (BCQ +43.8, MCQ +30.6) with open-ended gains (for example SD +21.2, VS +35.1) whose interpretation is not yet established. The manuscript itself concedes in Section 6 that BERTScore 'measures semantic similarity rather than logical correctness'. To support the central 'reasoning beyond detection' claim, I request an independent-reference check, a human evaluation of a sample of fine-tuned versus zero-shot outputs, or a reference-free/causal metric, and at minimum a separate reporting of closed-form and open-ended aggregate scores.
[§4.3, Table 4] The claim that 'training on more tasks improves untrained tasks' is based on improvements in Temporal Description, Causal Linkage, Scene Description, and Video Summarization, all of which are scored with BERTScore. In the QA-only configuration these tasks are not in the training set, but the model is trained on MAVEN-style open-ended answers from other QA tasks, so the observed gains could reflect transfer of stylistic conventions and output format rather than transfer of reasoning ability. Because this is a load-bearing part of the multi-task design argument, the manuscript should validate these specific gains with a second scoring method (independent references or human assessment) or temper the claim accordingly.

minor comments (4)

[Figure 1 caption] The figure caption says 'TAD and TAR-Bench provide multi-task reasoning annotations' but the paper elsewhere uses TAR for the training dataset; this appears to be a typo for 'TAR and TAR-Bench' and should be corrected.
[Table 3] The table title 'Zero-shot evaluation on TAR-Bench' is inaccurate for the Cosmos3 rows, since the paper notes that TAR training data was included in Cosmos3 base-model training; consider renaming the table or adding a clearer footnote to distinguish prior-exposure models from true zero-shot models.
[§3.5 and Appendix A] The human-correction statistics are useful, but the paper should state explicitly that the 0.95 median string similarity between original and corrected references means the TAR-Bench references remain highly similar to MAVEN output, which is directly relevant to the BERTScore interpretation discussed in the major comments.
[§4.4] The diagnostic error analysis reports percentages such as 18% anomaly omission and 30% zero-IoU temporal localization, but does not state how many examples were manually analyzed; adding the sample size would help readers judge the reliability of these estimates.

Circularity Check

0 steps flagged · score 2.0 of 10

No construction-level circularity: TAR-Bench is human-corrected and held out, and no prediction reduces to a fitted parameter or to a definitional identity. The MAVEN self-citation is descriptive, and the BERTScore style-matching concern is an acknowledged evaluation limitation rather than a circular derivation.

full rationale

Walking the paper's derivation chain, the central empirical claims are (i) strong QA accuracy does not predict temporal/scene reasoning, and (ii) progressive multi-task fine-tuning on TAR improves TAR-Bench scores. Neither claim is obtained by defining an output in terms of an input, by fitting a parameter and then renaming the fit as a prediction, or by importing a uniqueness theorem from the authors' prior work. The TAR-Bench evaluation set is held out from the training corpus: 'TAR-Bench contains 960 human-curated test annotations for 80 held-out clips trimmed from 17 public YouTube videos,' and 'Test annotations were drafted with MAVEN and then corrected by four expert reviewers.' The human-correction step provides a measure of independence from the training-label generator, and the paper discloses the residual overlap transparently. The open-ended tasks use BERTScore, and the paper itself concedes in Section 6 that 'BERTScore is deterministic and scalable but measures semantic similarity rather than logical correctness.' This is a real validity limitation for interpreting open-ended gains as gains in logical reasoning, but it is not a circular reduction: the fine-tuned model has not seen TAR-Bench answers, and BERTScore is not defined in terms of the training objective or the MAVEN output distribution. Closed-form tasks are scored by accuracy and temporal localization by mIoU, both independent of the annotation-style concern. The self-citation to MAVEN (Zhang et al., 2026) is used to describe the annotation pipeline, not to justify the benchmark's correctness or to forbid alternative explanations, so it is not load-bearing. No fitted parameter, definitional identity, or self-citation chain makes the reported gains equivalent to the inputs by construction. The appropriate finding is therefore no significant circularity, with the MAVEN self-citation and the shared annotation-style issue accounting for the low non-zero score.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

No fitted free parameters appear in the central results; the paper reports benchmark scores. The claims rest on the fidelity of the MAVEN annotation pipeline, the reliability of human-corrected TAR-Bench references, the adequacy of BERTScore as a reasoning metric, and the representativeness of the small YouTube-based test set.

assumptions (4)

domain assumption MAVEN's multi-scale captions and MSTED faithfully represent video content for annotation generation.
TAR's 44,040 training annotations are produced by MAVEN from Gemini 3.1 Pro and Gemma-4-31B outputs; if these captions misrepresent events, the training labels inherit errors. Invoked in Section 3.4.
domain assumption Human corrections on TAR-Bench yield a reliable reference standard despite high correction rates (17.7% of questions, 37.1% of answers).
The benchmark's validity rests on four expert reviewers; the paper reports substantive corrections but no inter-annotator agreement on the final references. See Section 3.5 and Appendix A.
domain assumption BERTScore is an adequate proxy for answer quality on open-ended reasoning tasks.
The paper acknowledges BERTScore measures semantic similarity rather than logical correctness (Section 6), yet uses it as the primary metric for seven of ten tasks.
domain assumption The 80 TAR-Bench clips from 17 YouTube videos are representative of the traffic anomaly reasoning distribution.
Results may have substantial source-level correlation; the paper recommends complementing with OOD evaluations (Section 6).

how reviews work

0 comments

Cite this review

Pith. "Pith review of From Detection to Understanding: TAR and TAR-Bench for Multi-Task Traffic Anomaly Reasoning." pith.science (2026). https://pith.science/paper/TQD6MYEY

@misc{pith2026260810317,
  author       = {Pith},
  title        = {Pith review of: From Detection to Understanding: TAR and TAR-Bench for Multi-Task Traffic Anomaly Reasoning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TQD6MYEY}},
  note         = {Machine review of arXiv:2608.10317}
}

abstract

We present TAR (Traffic Anomaly Reasoning) and TAR-Bench datasets, resources for training and evaluating video-language models beyond anomaly detection. TAR contains 44,040 chain-of-thought training annotations across 10 tasks for 3,670 CCTV videos ($\sim$26 hours) from eight public datasets. Its evaluation component, TAR-Bench, contains 960 human-curated test annotations for 80 held-out clips trimmed from 17 public YouTube videos. TAR's training annotations are produced with MAVEN, which consolidates multi-scale video evidence into structured event descriptions before generating question-answer pairs and reasoning traces. On TAR-Bench, eleven vision-language models reveal that strong question-answering accuracy does not reliably predict temporal or scene reasoning ability. Multi-task fine-tuning on TAR yields consistent gains, with the full 10-task model improving aggregate score by 21.4 points over its zero-shot baseline. TAR and TAR-Bench provide the official training and in-domain evaluation data for AI City Challenge 2026 Track 3. The dataset is available at https://huggingface.co/datasets/nvidia/PhysicalAI-Traffic-Anomaly-Reasoning

Figures

Figures reproduced from arXiv: 2608.10317 by the authors.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 23 canonical work pages

[1]

Cosmos 3: Omnimodal world models for physical ai.arXiv preprint arXiv:2606.02800, 2026

Niket Agarwal, Arslan Ali, Jon Allen, Martin Antolini, Adeline Aubame, Alisson Azzolini, Junjie Bai, Maciej Bala, Yogesh Balaji, Josh Bapst, et al. Cosmos 3: Omnimodal world models for physical ai.arXiv preprint arXiv:2606.02800, 2026. 6

arXiv 2026
[2]

AI City Challenge 2026 Track 3: Anomalous events in transportation

AI City Challenge. AI City Challenge 2026 Track 3: Anomalous events in transportation. https: //www.aicitychallenge.org/2026-track3/, 2026. 2, 9

work page 2026
[3]

Anastasiu

David C. Anastasiu. Explainable AI for real-time video anomaly anticipation. InProceedings of the SIAM International Conference on Data Mining (SDM), Blue Sky Track, 2025. 2

work page 2025
[4]

Highway traffic videos dataset.https://www.kaggle.com/datasets/aryashah2k/highw ay-traffic-videos-dataset, 2023

Aryashah2k. Highway traffic videos dataset.https://www.kaggle.com/datasets/aryashah2k/highw ay-traffic-videos-dataset, 2023. CC0 license. Washington State DOT source. 4

work page 2023
[5]

SO-TAD: A surveillance-oriented benchmark for traffic accident detection.Neurocomputing, 618:129061, 2025

Xingyuan Chen, Huahu Xu, Mingyang Ruan, Minjie Bian, Qishen Chen, and Yuzhe Huang. SO-TAD: A surveillance-oriented benchmark for traffic accident detection.Neurocomputing, 618:129061, 2025. 3, 4

work page 2025
[6]

CARLA: An open urban driving simulator

Alexey Dosovitskiy, German Ros, Felipe Codevilla, Antonio Lopez, and Vladlen Koltun. CARLA: An open urban driving simulator. InConference on Robot Learning (CoRL), pages 1–16, 2017. 9

work page 2017
[7]

VLMEvalKit: An open-source toolkit for evaluating large multi-modality models

Haodong Duan, Junming Yang, Yuxuan Qiao, Xinyu Fang, Lin Chen, Yuan Liu, Xiaoyi Dong, Yuhang Zang, Pan Zhang, Jiaqi Wang, et al. VLMEvalKit: An open-source toolkit for evaluating large multi-modality models. InProceedings of the 32nd ACM International Conference on Multimedia, pages 11198–11201,

work page
[8]

FishEye8K: A benchmark and dataset for fisheye camera object detection

Munkhjargal Gochoo, Munkh-Erdene Otgonbold, Erkhembayar Ganbold, Jun-Wei Hsieh, Ming-Ching Chang, Ping-Yang Chen, Byambaa Dorj, Hamad Al Jassmi, Ganzorig Batnasan, Fady Alnajjar, Mohammed Abduljabbar, and Fang-Pang Lin. FishEye8K: A benchmark and dataset for fisheye camera object detection. InProceedings of the IEEE/CVF Conference on Computer Vision and P...

work page 2023

Show all 34 references

[9]

FETV: Fisheye traffic violation dataset.https://gith ub.com/MoyoG/FETV, 2026

Munkhjargal Gochoo, Ahmed Abduljawad, et al. FETV: Fisheye traffic violation dataset.https://gith ub.com/MoyoG/FETV, 2026. 9

2026
[10]

Gemini 3 model documentation.https://ai.google.dev/gemini-api/docs/mode ls, 2026

Google DeepMind. Gemini 3 model documentation.https://ai.google.dev/gemini-api/docs/mode ls, 2026. 4, 6

2026
[11]

Gemma 4 model card.https://ai.google.dev/gemma/docs/core/model_card_4,

Google DeepMind. Gemma 4 model card.https://ai.google.dev/gemma/docs/core/model_card_4,
[12]

AccidentBench: Benchmarkingmultimodal understanding and reasoning in vehicle accidents and beyond.arXiv preprint arXiv:2509.26636, 2025

Shangding Gu, Xiaohan Wang, Donghao Ying, Haoyu Zhao, Runing Yang, Ming Jin, Boyi Li, Marco Pavone, SerenaYeung-Levy, JunWang, DawnSong, andCostasSpanos. AccidentBench: Benchmarkingmultimodal understanding and reasoning in vehicle accidents and beyond.arXiv preprint arXiv:2509...

2025
[13]

Vad- R1: Towards video anomaly reasoning via perception-to-cognition chain-of-thought.arXiv preprint arXiv:2505.19877, 2025

Chao Huang, Benfeng Wang, Jie Wen, Chengliang Liu, Wei Wang, Li Shen, and Xiaochun Cao. Vad- R1: Towards video anomaly reasoning via perception-to-cognition chain-of-thought.arXiv preprint arXiv:2505.19877, 2025. 3, 4, 6

2025 arXiv
[14]

Psi: A benchmark for human interpretation and response in traffic interactions

Taotao Jing, Tina Chen, Renran Tian, Yaobin Chen, Joshua Domeyer, Heishiro Toyoda, Rini Sherony, and Zhengming Ding. Psi: A benchmark for human interpretation and response in traffic interactions. In Advances in Neural Information Processing Systems, volume 38, 2025. 9 13 From...

2025
[15]

Gonzalez, Hao Zhang, and Ion Stoica

Woosuk Kwon, Zhuohan Li, Siyuan Zhuang, Ying Sheng, Lianmin Zheng, Cody Hao Yu, Joseph E. Gonzalez, Hao Zhang, and Ion Stoica. Efficient memory management for large language model serving with PagedAttention. InProceedings of the 29th Symposium on Operating Systems Principles ...

2023
[16]

Abnormal event detection at 150 FPS in MATLAB

Cewu Lu, Jianping Shi, and Jiaya Jia. Abnormal event detection at 150 FPS in MATLAB. InProceedings of the IEEE International Conference on Computer Vision (ICCV), pages 2720–2727, 2013. 3

2013
[17]

A revisit of sparse coding based anomaly detection in stacked RNN framework

Weixin Luo, Wen Liu, and Shenghua Gao. A revisit of sparse coding based anomaly detection in stacked RNN framework. InProceedings of the IEEE International Conference on Computer Vision (ICCV), pages 341–349, 2017. 2, 3

2017
[18]

Localizing anomalies from weakly-labeled videos.IEEE Transactions on Image Processing, 2021

Hui Lv, Chuanwei Zhou, Zhen Cui, Chunyan Xu, Yong Li, and Jian Yang. Localizing anomalies from weakly-labeled videos.IEEE Transactions on Image Processing, 2021. 2, 3, 4

2021
[19]

Cosmos-Reason2-8B.https://huggingface.co/nvidia/Cosmos-Reason2-8B, 2026

NVIDIA. Cosmos-Reason2-8B.https://huggingface.co/nvidia/Cosmos-Reason2-8B, 2026. 6

2026
[20]

Cosmos-Reason2-32B.https://huggingface.co/nvidia/Cosmos-Reason2-32B, 2026

NVIDIA. Cosmos-Reason2-32B.https://huggingface.co/nvidia/Cosmos-Reason2-32B, 2026. 6

2026
[21]

Qwen3-VL technical report.arXiv preprint arXiv:2511.21631, 2025

Qwen Team. Qwen3-VL technical report.arXiv preprint arXiv:2511.21631, 2025. 6

2025 arXiv
[22]

Shah, Jean-Baptiste Lamare, Tuan Nguyen-Anh, and Alexander Hauptmann

Ankit P. Shah, Jean-Baptiste Lamare, Tuan Nguyen-Anh, and Alexander Hauptmann. CADP: A novel dataset for CCTV traffic camera based accident analysis.arXiv preprint arXiv:1809.05782, 2018. 3

2018 arXiv
[23]

Real-world anomaly detection in surveillance videos

Waqas Sultani, Chen Chen, and Mubarak Shah. Real-world anomaly detection in surveillance videos. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 6479–6488,
[24]

Zheng Tang, Shuo Wang, David C. Anastasiu, Ming-Ching Chang, Anuj Sharma, Quan Kong, Norimasa Kobori, Munkhjargal Gochoo, Ganzorig Batnasan, Munkh-Erdene Otgonbold, Fady Alnajjar, Jun-Wei Hsieh, Tomasz Kornuta, Xiaolong Li, Yilin Zhao, Han Zhang, Subhashree Radhakrishnan, Arih...

2025 arXiv
[25]

Anastasiu, Ming-Ching Chang, et al

Zheng Tang, Shuo Wang, David C. Anastasiu, Ming-Ching Chang, et al. The 10th AI City Challenge. In ECCV Workshops, Malm"o, Sweden, 2026. 2, 9

2026
[26]

PSI VQA: Egocentric dashcam pedestrian intent benchmark.https: //huggingface.co/datasets/ise-ice-lab/PSI_VQA, 2026

Renran Tian, Shaozhi Wang, et al. PSI VQA: Egocentric dashcam pedestrian intent benchmark.https: //huggingface.co/datasets/ise-ice-lab/PSI_VQA, 2026. 9

2026
[27]

TAD: A large-scale benchmark for traffic accidents detection from video surveillance.arXiv preprint arXiv:2209.12386, 2022

Yajun Xu, Chuwen Huang, Yibing Nan, and Shiguo Lian. TAD: A large-scale benchmark for traffic accidents detection from video surveillance.arXiv preprint arXiv:2209.12386, 2022. 3, 4

2022 arXiv
[28]

MAVEN: A multi-stage agentic annotation pipeline for video reasoning tasks.arXiv preprint arXiv:2605.21917, 2026

Han Zhang, Wanting Jiang, Tomasz Kornuta, Tian Zheng, and Vidya Murali. MAVEN: A multi-stage agentic annotation pipeline for video reasoning tasks.arXiv preprint arXiv:2605.21917, 2026. URL https://arxiv.org/abs/2605.21917. 3, 4, 9

2026 arXiv
[29]

Holmes-VAD: Towards unbiased and explainable video anomaly detection via multi-modal LLM.arXiv preprint arXiv:2406.12235, 2024

Huaxin Zhang, Xiaohao Xu, Xiang Wang, Jialong Zuo, Chuchu Han, Xiaonan Huang, Changxin Gao, Yuehuan Wang, and Nong Sang. Holmes-VAD: Towards unbiased and explainable video anomaly detection via multi-modal LLM.arXiv preprint arXiv:2406.12235, 2024. URLhttps://arxiv.org/abs/2406.122

2024 arXiv
[30]

Holmes-VAU: Towards long-term video anomaly understanding at any granularity

Huaxin Zhang, Xiaohao Xu, Xiang Wang, Jialong Zuo, Xiaonan Huang, Changxin Gao, Shanjun Zhang, Li Yu, and Nong Sang. Holmes-VAU: Towards long-term video anomaly understanding at any granularity. arXiv preprint arXiv:2412.06171, 2024. URLhttps://arxiv.org/abs/2412.06171. 3 14 F...

2024 arXiv
[31]

Weinberger, and Yoav Artzi

Tianyi Zhang, Varsha Kishore, Felix Wu, Kilian Q. Weinberger, and Yoav Artzi. BERTScore: Evaluating text generation with BERT. InInternational Conference on Learning Representations (ICLR), 2020. 6

2020
[32]

SurveillanceVQA-589K: A benchmark for comprehensive surveillance video-language understanding with large models.arXiv preprint arXiv:2505.12589, 2025

Yuqi Zhang et al. SurveillanceVQA-589K: A benchmark for comprehensive surveillance video-language understanding with large models.arXiv preprint arXiv:2505.12589, 2025. 3

2025 arXiv
[33]

Towards surveillance video-and-language understanding: New dataset, baselines, and challenges.arXiv preprint arXiv:2309.13925, 2023

Tongtong Zhao et al. Towards surveillance video-and-language understanding: New dataset, baselines, and challenges.arXiv preprint arXiv:2309.13925, 2023. 3

2023 arXiv
[34]

Barbados traffic analysis challenge.https://zindi.africa/competitions/barbados-traffic -analysis-challenge/data, 2023

Zindi. Barbados traffic analysis challenge.https://zindi.africa/competitions/barbados-traffic -analysis-challenge/data, 2023. 4 15

2023

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.

[1] [1]

Cosmos 3: Omnimodal world models for physical ai.arXiv preprint arXiv:2606.02800, 2026

Niket Agarwal, Arslan Ali, Jon Allen, Martin Antolini, Adeline Aubame, Alisson Azzolini, Junjie Bai, Maciej Bala, Yogesh Balaji, Josh Bapst, et al. Cosmos 3: Omnimodal world models for physical ai.arXiv preprint arXiv:2606.02800, 2026. 6

arXiv 2026

[2] [2]

AI City Challenge 2026 Track 3: Anomalous events in transportation

AI City Challenge. AI City Challenge 2026 Track 3: Anomalous events in transportation. https: //www.aicitychallenge.org/2026-track3/, 2026. 2, 9

work page 2026

[3] [3]

Anastasiu

David C. Anastasiu. Explainable AI for real-time video anomaly anticipation. InProceedings of the SIAM International Conference on Data Mining (SDM), Blue Sky Track, 2025. 2

work page 2025

[4] [4]

Highway traffic videos dataset.https://www.kaggle.com/datasets/aryashah2k/highw ay-traffic-videos-dataset, 2023

Aryashah2k. Highway traffic videos dataset.https://www.kaggle.com/datasets/aryashah2k/highw ay-traffic-videos-dataset, 2023. CC0 license. Washington State DOT source. 4

work page 2023

[5] [5]

SO-TAD: A surveillance-oriented benchmark for traffic accident detection.Neurocomputing, 618:129061, 2025

Xingyuan Chen, Huahu Xu, Mingyang Ruan, Minjie Bian, Qishen Chen, and Yuzhe Huang. SO-TAD: A surveillance-oriented benchmark for traffic accident detection.Neurocomputing, 618:129061, 2025. 3, 4

work page 2025

[6] [6]

CARLA: An open urban driving simulator

Alexey Dosovitskiy, German Ros, Felipe Codevilla, Antonio Lopez, and Vladlen Koltun. CARLA: An open urban driving simulator. InConference on Robot Learning (CoRL), pages 1–16, 2017. 9

work page 2017

[7] [7]

VLMEvalKit: An open-source toolkit for evaluating large multi-modality models

Haodong Duan, Junming Yang, Yuxuan Qiao, Xinyu Fang, Lin Chen, Yuan Liu, Xiaoyi Dong, Yuhang Zang, Pan Zhang, Jiaqi Wang, et al. VLMEvalKit: An open-source toolkit for evaluating large multi-modality models. InProceedings of the 32nd ACM International Conference on Multimedia, pages 11198–11201,

work page

[8] [8]

FishEye8K: A benchmark and dataset for fisheye camera object detection

Munkhjargal Gochoo, Munkh-Erdene Otgonbold, Erkhembayar Ganbold, Jun-Wei Hsieh, Ming-Ching Chang, Ping-Yang Chen, Byambaa Dorj, Hamad Al Jassmi, Ganzorig Batnasan, Fady Alnajjar, Mohammed Abduljabbar, and Fang-Pang Lin. FishEye8K: A benchmark and dataset for fisheye camera object detection. InProceedings of the IEEE/CVF Conference on Computer Vision and P...

work page 2023