Pith. sign in

REVIEW 4 major objections 6 minor 58 references

VRU-Accident: A Vision-Language Benchmark for Video Question Answering and Dense Captioning for Accident Scene Understanding

T0 review · 4 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read AI models lag humans by 28 points on pedestrian-crash reasoning

desk verdict A useful VRU accident benchmark that should get peer review, but the zero-shot numbers need a source-split re-evaluation before the performance gap can be taken at face value. read the letter →

arxiv 2507.09815 v2 pith:OH2LCDXR submitted 2025-07-13 cs.CV

Younggun Kim , Ahmed S. Abdelrahman , Mohamed Abdel-Aty This is my paper
classification cs.CV
keywords videoquestionansweringdensecaptioningvulnerableroadusersmultimodallargelanguagemodelsaccidentsceneunderstandingautonomousdrivingsafetycounterfactualdistractorsbenchmark
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

VRU-Accident is a large-scale vision-language benchmark built from 1,000 real-world dashcam videos of accidents and near-accidents involving pedestrians and cyclists, annotated with 6,000 multiple-choice questions across six safety-critical categories and 1,000 dense scene descriptions. The paper's central claim is that current multimodal large language models (MLLMs) can answer visually grounded questions about weather, lighting, and traffic environment but cannot reliably reason about accident cause, type, and prevention. In zero-shot evaluation of 17 models, the best model reaches 66.9% average accuracy against 94.7% for human experts, and dense captions frequently omit collision dynamics or hallucinate road-user motion. The benchmark is offered as a standardized testbed for diagnosing and improving accident-centric reasoning in MLLMs for autonomous driving safety.

What carries the argument

The load-bearing mechanism is the benchmark's six-category VQA design paired with counterfactual distractor generation. For each video $V_i$ and each category $j$, human experts supply a ground-truth answer $y_j$; a VQA generator $G_{VQA}$ produces three contextually plausible but incorrect answers $y^*_j$, so the candidate set $\hat{Y}_j = \{y_j, y^{*(1)}_j, y^{*(2)}_j, y^{*(3)}_j\}$ forces models to reason over dynamically constructed, semantically diverse answer spaces rather than choose from a fixed label set. A second generator $G_{DC}$ produces dense captions $C_i$ describing weather, environment, road-user appearance, kinematics, spatial relations, and collision sequence, and human experts verify all annotations. The three causal categories — Accident Type, Accident Cause, and Prevention Measure — carry the diagnostic weight: they separate visually grounded attribute recognition from the causal and counterfactual reasoning that the paper claims current MLLMs lack.

What would settle it

Run the same 17 models on a newly recorded set of 1,000 VRU accident and near-accident videos that were never posted publicly and postdate every model's training cutoff, using the identical VQA and dense-caption protocol. If average accuracy approaches the 94.7% human level, the claimed reasoning deficit is contamination, not a genuine capability gap.

Watch

Extended reading notes

Core claim

The paper introduces VRU-Accident as the first large-scale vision-language benchmark aimed specifically at evaluating MLLMs in safety-critical scenarios involving vulnerable road users. Each of its 1,000 ego-view dashcam videos carries six VQA pairs, one per category — Weather & Light, Traffic Environment, Road Configuration, Accident Type, Accident Cause, and Accident Prevention Measure — with one correct answer and three counterfactual distractors per question, yielding 24,000 candidate options and over 3.4K unique answers; each video also has a dense, temporally grounded caption. Annotations are produced semi-automatically: human experts label ground truth, GPT-4o generates plausible distractors and initial dense captions, and human experts verify. Evaluated zero-shot on 17 MLLMs (15 open-source and 2 closed-source), the benchmark shows that models do well on visually grounded attributes but drop sharply on causal and preventive reasoning: the best model, Gemini 1.5-flash, averages 66.9% versus 94.7% for human experts, and dense captions score low on SPICE, METEOR, COMET, and ROUGE while often misdescribing collision dynamics. The authors conclude that VRU-Accident provides a diagnostic and generative testbed that exposes the gap between visual grounding and true accident reasoning in current MLLMs.

Load-bearing premise

The conclusion that today's MLLMs lack VRU accident reasoning assumes the reused public dashcam videos were not part of the models' pretraining data; if the models have already seen these clips, the reported zero-shot accuracy gaps are inflated.

Editorial extensions

If this is right

  • Model rankings on VRU-Accident give a standardized zero-shot measure of accident-scene reasoning, so progress in safety-critical MLLMs can be tracked against a fixed human baseline of 94.7%.
  • Because visually grounded categories are easy while Accident Cause and Prevention Measure are hard, benchmarks that only test scene attributes will overstate MLLM readiness for autonomous driving perception.
  • The counterfactual distractor design means a model must distinguish the correct cause from plausible alternatives rather than memorize label distributions, so high accuracy requires genuine causal reasoning.
  • Low dense-caption scores and hallucinated pedestrian motion show that current MLLMs lack temporally grounded collision understanding, not just vocabulary for describing scenes.
  • Open-source models such as Qwen2-VL(7B) can serve as practical baselines for accident captioning, but all evaluated models fall short of human-level narrative accuracy.

Reading between the lines

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

  • Beyond the paper, because the videos are drawn from public datasets MM-AU and DoTA, an implicit assumption is that the evaluated MLLMs did not see these exact clips during pretraining; a clean held-out test with newly collected videos would be a stronger test of the claimed reasoning gap.
  • Beyond the paper, the counterfactual distractor pairs could be reused as training data for contrastive accident reasoning, teaching models to prefer causally correct explanations over visually plausible ones.
  • Beyond the paper, reference captions were initially generated by GPT-4o, so reported captioning scores may understate models whose narrative style differs from GPT-4o; rewriting references by human experts would harden the dense-caption leaderboard.
  • Beyond the paper, the six categories could be extended to near-miss severity, VRU intention prediction, and liability attribution, which would make the benchmark directly useful for autonomous vehicle decision-making rather than only for diagnosis.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. VRU-Accident introduces a benchmark of 1,000 real-world dashcam videos of accidents or near-accidents involving vulnerable road users, annotated with 6,000 multiple-choice VQA pairs across six categories (24,000 candidate options, 3,400+ unique answers) and 1,000 dense scene descriptions. The curation pipeline is semi-automatic: human experts provide correct VQA answers, GPT-4o generates counterfactual distractors and initial dense captions, and human experts verify the results. The paper evaluates 17 MLLMs (15 open-source, 2 closed-source) in a zero-shot setting on VQA and dense captioning, reporting best average VQA accuracy of 66.9% (Gemini 1.5-flash) versus 94.7% for human experts, with the largest gaps on Accident Cause and Prevention Measure. The central conclusion is that current MLLMs handle visually grounded attributes reasonably well but struggle with causal, preventive, and fine-grained accident reasoning.

Significance. If the benchmark numbers are taken at face value, VRU-Accident is a useful and timely resource: it targets an under-served safety-critical niche, combines VQA and dense captioning in one benchmark, includes human verification in the annotation loop, covers diverse VRU accident scenarios, and ships inference scripts, model outputs, and evaluation code for reproducibility. The paper's qualitative finding that all evaluated models fall far below human performance on causal and preventive reasoning is plausible and worth publishing, provided the validity concerns below are resolved. The benchmark's potential impact for MLLM evaluation in autonomous driving justifies serious attention, but the headline numbers currently rest on assumptions about video contamination and reference independence that are not verified.

major comments (4)
  1. [Section 2, Table 1; Section 4.1] The paper states in Section 2 that VRU-Accident 'combines the MM-AU and DoTA samples while adding more samples reaching 1,000 videos.' MM-AU and DoTA are public web-scraped dashcam datasets, and the evaluated MLLMs were pretrained on large web-scale video corpora. The zero-shot protocol described in Section 4.1 therefore does not guarantee that the videos are unseen, and the Table 4 accuracies may conflate genuine reasoning with memorization of specific footage. This is load-bearing for the central claim that models 'perform reasonably well on visually grounded attributes' yet fail on causal reasoning. The authors should provide a source-split analysis (e.g., report results separately for newly collected videos versus reused MM-AU/DoTA videos) and a decontamination check, or explicitly weaken the zero-shot claim. Decontaminated scores would likely be lower, not higher, so the qualitative conclusion about causal reasoning may survive, but the benchmark's quantitative claims cannot be accepted without this analysis.
  2. [Section 3.3, Supplementary D, Table 5] The dense captioning ground truth is generated by GPT-4o and then human-verified, rather than written independently by human annotators. The paper appropriately excludes GPT-4 from the dense captioning leaderboard in Table 5, but all other models are still scored against references produced by the same model family and in the same narrative style. As a result, the reported SPICE, METEOR, COMET, and ROUGE scores reward stylistic and content choices of GPT-4o, not necessarily accident-understanding quality. This is load-bearing for the dense captioning conclusions. The authors should either collect a human-written reference subset, add human preference or correctness evaluation, or report robustness of model rankings across alternative reference sets.
  3. [Section 4.2, Table 4] The human expert row in Table 4 reports 94.7% average accuracy, but the manuscript does not state the number of annotators, their qualifications, the annotation protocol, or inter-annotator agreement. Without this information, the human baseline and the label reliability of the VQA ground truth are not established. This matters because the paper's central comparison is between MLLM performance and human expert performance, and because the benchmark's utility depends on the correctness and consistency of the ground-truth answers. The authors should report annotation statistics and agreement metrics, and describe how borderline cases (e.g., near-miss events discussed in Supplementary B.3) were adjudicated.
  4. [Section 3.2, Supplementary D] The VQA distractors are generated by GPT-4o from the human-labeled ground truth, and the analysis in Supplementary Figure 8 shows that distractors in several categories have question-to-answer cosine similarity comparable to or higher than the ground truth. While this suggests the distractors are semantically plausible, it also raises the question of whether some multiple-choice items are ambiguous or have more than one defensible answer. The paper does not report human agreement on the final candidate sets or any validation that the three distractors are unambiguously incorrect. Since the benchmark's discriminative power depends on distractor quality, the authors should provide human validation statistics or a discussion of ambiguity resolution.
minor comments (6)
  1. [Throughout] There are numerous typos and formatting artifacts, including 'traffic accidentes' in the Introduction, 'One the other hand' in Section 2, 'Unlikely accident classification tasks' in Section 3.4 (should be 'Unlike'), 'Configiguration' in Table 3, and inconsistent rendering of 'LLaVA' as 'LLaV A' throughout the text and tables.
  2. [Table 4 and Figure 4] The model lists in Table 4 and Figure 4 are inconsistent: Figure 4 includes 'Mobile-VideoGPT(0.5B)', which does not appear in Table 4, while Table 4 lists 'Mobile-VideoGPT(1.5B)'. The authors should align these lists and describe which model versions were actually evaluated.
  3. [Supplementary B.2, Eq. (4)] The METEOR formula as written, METEOR = (Precision × Recall) / (Precision + α × Recall + (1 − α)), does not match the standard METEOR formulation in the cited reference [19]. Please clarify the formula or use the standard definition with the appropriate penalty term.
  4. [Section 3.3 and Supplementary D] The main text refers to a 'Dense Captioning generator G_DC' without identifying it as GPT-4o; the identity is only revealed in the supplementary material. This should be stated explicitly in the main text, since it directly affects interpretation of the dense captioning evaluation.
  5. [Section 4.1] The paper says BLEU is omitted because scores are below 0.1, but BLEU is a precision-oriented metric that is known to be harsh on video captions; a one-sentence justification in the main text would help readers interpret the metric choice.
  6. [Section 2 and Data Release] Since the benchmark reuses videos from MM-AU and DoTA, the paper should document the licenses and any privacy or consent considerations for the dashcam footage, especially because the dataset is publicly released and contains identifiable people in accident situations.

Circularity Check

1 steps flagged · score 2.0 of 10

Minor circularity: dense captioning references are generated by GPT-4o, but the central VQA evaluation is human-labeled and independent.

  1. self definitional [Section 3.3 (Dense Caption Curation Pipeline) and Section 4.2 (Dense Captioning Task)]
    ""we employ a Dense Captioning generator GDC that produces a caption Ci for each video Vi, such that Ci = GDC(Vi)" and "GPT-4 is excluded from this comparison, as it was used to generate the initial reference captions for the videos in the VRU-Accident benchmark.""

    The dense captioning 'ground truth' is itself the output of an MLLM (GPT-4o), so every SPICE, METEOR, COMET, and ROUGE score is by construction a similarity measure between the tested model and GPT-4o's narrative style, not an independently human-authored reference. This makes the dense captioning leaderboard partially self-referential. The paper mitigates this by excluding GPT-4 from the comparison, and the VQA ground-truth answers are human-labeled with only the distractor options generated by GPT-4o, so the main VQA finding does not reduce to this step.

full rationale

The paper's central claim is the VQA evaluation of 17 MLLMs against 6K human-labeled question-answer pairs; that result is independent of any model-generated reference and is not circular. The only self-referential element is the dense captioning task, where the reference captions are produced by GPT-4o and then used as the evaluation target; this is disclosed and partially mitigated by excluding GPT-4 from the dense captioning leaderboard. Self-citations in the paper are contextual or empirical and are not load-bearing for the benchmark's conclusions. The concern that reused MM-AU/DoTA videos may overlap with MLLM pretraining corpora is a data-contamination threat to external validity, not a circularity in the paper's derivation, so it does not raise the circularity score.

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

The benchmark introduces no new theoretical entities or fitted parameters. The load-bearing assumptions are about annotation reliability, distractor quality, video representativeness, and the validity of using GPT-4-generated captions as ground truth for dense captioning evaluation.

assumptions (4)
  • domain assumption Human experts' manual labels for the VQA ground truth are correct and consistent.
    The benchmark assumes the expert annotations used to compute ground truth and human accuracy are reliable; the paper does not report inter-annotator agreement.
  • domain assumption GPT-4o-generated counterfactual distractors are plausible but incorrect and do not introduce systematic ambiguity.
    The VQA difficulty relies on the distractors being semantically plausible; if distractors are easily distinguishable or sometimes arguably correct, accuracy numbers are distorted.
  • domain assumption The 1,000 videos from web searches and existing datasets are representative of VRU accidents.
    The benchmark's conclusions about MLLMs' accident-reasoning capability assume the video set is representative; manual filtering may bias the sample.
  • domain assumption GPT-4-generated dense caption references are high-quality ground truth for evaluating other models.
    Dense captioning evaluation compares models against GPT-4-generated captions; if these captions are stylistically biased, the leaderboard measures alignment with GPT-4 rather than human performance.

how reviews work

0 comments
Cite this review

Pith. "Pith review of VRU-Accident: A Vision-Language Benchmark for Video Question Answering and Dense Captioning for Accident Scene Understanding." pith.science (2026). https://pith.science/paper/OH2LCDXR

@misc{pith2026250709815,
  author       = {Pith},
  title        = {Pith review of: VRU-Accident: A Vision-Language Benchmark for Video Question Answering and Dense Captioning for Accident Scene Understanding},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OH2LCDXR}},
  note         = {Machine review of arXiv:2507.09815}
}
read the original abstract

Ensuring the safety of vulnerable road users (VRUs), such as pedestrians and cyclists, is a critical challenge for autonomous driving systems, as crashes involving VRUs often result in severe or fatal consequences. While multimodal large language models (MLLMs) have shown promise in enhancing scene understanding and decision making in autonomous vehicles, there is currently no standardized benchmark to quantitatively evaluate their reasoning abilities in complex, safety-critical scenarios involving VRUs. To address this gap, we present VRU-Accident, a large-scale vision-language benchmark designed to evaluate MLLMs in high-risk traffic scenarios involving VRUs. VRU-Accident comprises 1K real-world dashcam accident videos, annotated with 6K multiple-choice question-answer pairs across six safety-critical categories (with 24K candidate options and 3.4K unique answer choices), as well as 1K dense scene descriptions. Unlike prior works, our benchmark focuses explicitly on VRU-vehicle accidents, providing rich, fine-grained annotations that capture both spatial-temporal dynamics and causal semantics of accidents. To assess the current landscape of MLLMs, we conduct a comprehensive evaluation of 17 state-of-the-art models on the multiple-choice VQA task and on the dense captioning task. Our findings reveal that while MLLMs perform reasonably well on visually grounded attributes, they face significant challenges in reasoning and describing accident causes, types, and preventability.

Figures

Figures reproduced from arXiv: 2507.09815 by the authors.

Figure 1
Figure 1. VQA distribution in the VRU-Accident benchmark. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of the VRU-Accident benchmark curation pipeline. Accident videos are retrieved from web sources based on traffic [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Visualization of annotations in VRU-Accident benchmark illustrating both the VQA and dense captioning tasks. [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Summary of MLLMs performance on VRU-Accident [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Accident description from both Ground Truth in VRU-Accident and Qwen2-VL(7B) which achieved the best performance in the [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Word frequency per each category in the VRU-Accident benchmark. [PITH_FULL_IMAGE:figures/full_fig_p014_6.png]
Figure 8
Figure 8. Figure 8: Comparison of average Q-A cosine similarity for ground [PITH_FULL_IMAGE:figures/full_fig_p015_8.png]
Figure 9
Figure 9. Figure 9: Qualitative comparison of VQA and dense captioning outputs from MLLMs on the VRU-Accident benchmark. Each example [PITH_FULL_IMAGE:figures/full_fig_p020_9.png]
Figure 10
Figure 10. Figure 10: Prompt used for curating VRU-Accident VQA benchmark, including category-specific rules and a generated multiple-choice [PITH_FULL_IMAGE:figures/full_fig_p021_10.png]
Figure 11
Figure 11. Figure 11: Prompt used for curating dense caption annotations in VRU-Accident. The prompt guides the model to describe accident videos [PITH_FULL_IMAGE:figures/full_fig_p022_11.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

58 extracted references · 47 canonical work pages

  1. [1]

    Vru-cipi: Crossing intention prediction at intersections for improving vulnerable road users safety

    Ahmed S Abdelrahman, Mohamed Abdel-Aty, and Quoc Dai Tran. Vru-cipi: Crossing intention prediction at intersections for improving vulnerable road users safety. In Proceedings of the Computer Vision and Pattern Recognition Conference, pages 5623–5632, 2025. 1

    work page 2025

  2. [2]

    Video-to-text pedestrian monitoring (vtpm): Leveraging large language models for privacy-preserve pedestrian activity monitoring at intersections

    Ahmed S Abdelrahman, Mohamed Abdel-Aty, and Dong- dong Wang. Video-to-text pedestrian monitoring (vtpm): Leveraging large language models for privacy-preserve pedestrian activity monitoring at intersections. In Proceed- ings of the Winter Conference on Applications of Computer Vision, pages 366–375, 2025. 1

    work page 2025

  3. [3]

    Advanced crash causation analysis for freeway safety: A large language model ap- proach to identifying key contributing factors.arXiv preprint arXiv:2505.09949, 2025

    Ahmed S Abdelrahman, Mohamed Abdel-Aty, Samgyu Yang, and Abdulrahman Faden. Advanced crash causation analysis for freeway safety: A large language model ap- proach to identifying key contributing factors.arXiv preprint arXiv:2505.09949, 2025. 1

    arXiv 2025

  4. [4]

    Vrucrosssafe for crossing intention prediction of vulnerable road users for improving safe crossing at in- tersections

    Ahmed S Abdelrahman, Zubayer Islam, and Mohamed Abdel-Aty. Vrucrosssafe for crossing intention prediction of vulnerable road users for improving safe crossing at in- tersections. npj Sustainable Mobility and Transport , 2(1): 201–15, 2025. 2

    work page 2025

  5. [5]

    Evaluating the safety impact of mid-block pedes- trian signals (mps)

    Md Jamil Ahsan, Mohamed Abdel-Aty, and Ahmed S Ab- delrahman. Evaluating the safety impact of mid-block pedes- trian signals (mps). Accident Analysis & Prevention , 210: 107847, 2025. 2

    work page 2025

  6. [6]

    Spice: Semantic propositional image cap- tion evaluation

    Peter Anderson, Basura Fernando, Mark Johnson, and Stephen Gould. Spice: Semantic propositional image cap- tion evaluation. In Computer Vision–ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, Octo- ber 11-14, 2016, Proceedings, Part V 14 , pages 382–398. Springer, 2016. 6, 12

    work page 2016

  7. [7]

    How far are we to gpt-4v? closing the gap to commercial multimodal models with open-source suites

    Zhe Chen, Weiyun Wang, Hao Tian, Shenglong Ye, Zhang- wei Gao, Erfei Cui, Wenwen Tong, Kongzhi Hu, Jiapeng Luo, Zheng Ma, et al. How far are we to gpt-4v? closing the gap to commercial multimodal models with open-source suites. arXiv preprint arXiv:2404.16821, 2024. 3, 6, 7, 15

    arXiv 2024

  8. [8]

    Ex- panding performance boundaries of open-source multimodal models with model, data, and test-time scaling, 2025

    Zhe Chen, Weiyun Wang, Yue Cao, Yangzhou Liu, Zhang- wei Gao, Erfei Cui, Jinguo Zhu, Shenglong Ye, Hao Tian, Zhaoyang Liu, Lixin Gu, Xuehui Wang, Qingyun Li, Yimin Ren, Zixuan Chen, Jiapeng Luo, Jiahao Wang, Tan Jiang, Bo Wang, Conghui He, Botian Shi, Xingcheng Zhang, Han Lv, Yi Wang, Wenqi Shao, Pei Chu, Zhongying Tu, Tong He, Zhiyong Wu, Huipeng Deng, J...

    work page 2025

Show all 58 references
  1. [9]

    Analysis of automatic evaluation metric on low-resourced language: Bertscore vs bleu score

    Goutam Datta, Nisheeth Joshi, and Kusum Gupta. Analysis of automatic evaluation metric on low-resourced language: Bertscore vs bleu score. In International Conference on Speech and Computer, pages 155–162. Springer, 2022. 6

    2022

  2. [10]

    Dada-2000: Can driving accident be pre- dicted by driver attention? analyzed by a benchmark, 2019

    Jianwu Fang, Dingxin Yan, Jiahuan Qiao, Jianru Xue, He Wang, and Sen Li. Dada-2000: Can driving accident be pre- dicted by driver attention? analyzed by a benchmark, 2019. 1, 2, 3, 4

    2000

  3. [11]

    Cognitive accident prediction in driving scenes: A multimodality benchmark, 2023

    Jianwu Fang, Lei-Lei Li, Kuan Yang, Zhedong Zheng, Jianru Xue, and Tat-Seng Chua. Cognitive accident prediction in driving scenes: A multimodality benchmark, 2023. 2, 3, 14

    2023

  4. [12]

    Abductive ego-view accident video understanding for safe driving per- ception, 2024

    Jianwu Fang, Lei lei Li, Junfei Zhou, Junbin Xiao, Hongkai Yu, Chen Lv, Jianru Xue, and Tat-Seng Chua. Abductive ego-view accident video understanding for safe driving per- ception, 2024. 1, 2, 3, 4, 14

    2024

  5. [13]

    Pedestrian traffic fatalities by state: 2019 preliminary data, 2020

    Governors Highway Safety Association (GHSA). Pedestrian traffic fatalities by state: 2019 preliminary data, 2020. 2

    2019

  6. [14]

    Multi-frame, lightweight & efficient vision-language models for question answering in autonomous driving, 2024

    Akshay Gopalkrishnan, Ross Greer, and Mohan Trivedi. Multi-frame, lightweight & efficient vision-language models for question answering in autonomous driving, 2024. 1

    2024

  7. [15]

    Cipf: Crossing intention prediction network based on feature fusion modules for improving pedestrian safety

    Je-Seok Ham, Dae Hoe Kim, NamKyo Jung, and Jinyoung Moon. Cipf: Crossing intention prediction network based on feature fusion modules for improving pedestrian safety. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3666–3675, 2023. 1

    2023

  8. [16]

    An attention-guided multistream feature fusion network for early localization of risky traffic agents in driving videos

    Muhammad Monjurul Karim, Zhaozheng Yin, and Ruwen Qin. An attention-guided multistream feature fusion network for early localization of risky traffic agents in driving videos. IEEE Transactions on Intelligent Vehicles, 9(1):1792–1803,

  9. [17]

    Textual explanations for self-driving ve- hicles

    Jinkyu Kim, Anna Rohrbach, Trevor Darrell, John Canny, and Zeynep Akata. Textual explanations for self-driving ve- hicles. In Proceedings of the European conference on com- puter vision (ECCV), pages 563–578, 2018. 2

    2018

  10. [18]

    Pedes- trian crossing direction prediction at intersections for pedes- trian safety

    Younggun Kim, Mohamed Abdel-Aty, Keechoo Choi, Zubayer Islam, Dongdong Wang, and Shaoyan Zhai. Pedes- trian crossing direction prediction at intersections for pedes- trian safety. IEEE Open Journal of Intelligent Transportation Systems, 6:692–707, 2025. 2

    2025

  11. [19]

    Meteor: an automatic met- ric for mt evaluation with high levels of correlation with hu- man judgments

    Alon Lavie and Abhaya Agarwal. Meteor: an automatic met- ric for mt evaluation with high levels of correlation with hu- man judgments. In Proceedings of the Second Workshop on Statistical Machine Translation, page 228–231, USA, 2007. Association for Computational Linguistics. 6, 12

    2007

  12. [20]

    Llava-onevision: Easy visual task transfer,

    Bo Li, Yuanhan Zhang, Dong Guo, Renrui Zhang, Feng Li, Hao Zhang, Kaichen Zhang, Yanwei Li, Ziwei Liu, and Chunyuan Li. Llava-onevision: Easy visual task transfer,

  13. [21]

    Llava-next-interleave: Tackling multi-image, video, and 3d in large multimodal models, 2024

    Feng Li, Renrui Zhang, Hao Zhang, Yuanhan Zhang, Bo Li, Wei Li, Zejun Ma, and Chunyuan Li. Llava-next-interleave: Tackling multi-image, video, and 3d in large multimodal models, 2024. 3, 6, 7, 15

    2024

  14. [22]

    ROUGE: A package for automatic evaluation of summaries

    Chin-Yew Lin. ROUGE: A package for automatic evaluation of summaries. In Text Summarization Branches Out, pages 74–81, Barcelona, Spain, 2004. Association for Computa- tional Linguistics. 6, 12

    2004

  15. [23]

    Aligning llm with human travel choices: a persona-based embedding learning approach, 2025

    Tianming Liu, Manzi Li, and Yafeng Yin. Aligning llm with human travel choices: a persona-based embedding learning approach, 2025. 1

    2025

  16. [24]

    Toward llm- agent-based modeling of transportation systems: A concep- tual framework

    Tianming Liu, Jirong Yang, and Yafeng Yin. Toward llm- agent-based modeling of transportation systems: A concep- tual framework. Artificial Intelligence for Transportation, 1: 100001, 2025. 1

    2025

  17. [25]

    Video-xl-pro: Reconstructive token compression for extremely long video understanding, 2025

    Xiangrui Liu, Yan Shu, Zheng Liu, Ao Li, Yang Tian, and Bo Zhao. Video-xl-pro: Reconstructive token compression for extremely long video understanding, 2025. 3, 6, 7, 15

    2025

  18. [26]

    A simulation-based frame- work for urban traffic accident detection

    Haohan Luo and Feng Wang. A simulation-based frame- work for urban traffic accident detection. In ICASSP 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 1–5, 2023. 2, 3

    2023

  19. [27]

    Dolphins: Multimodal language model for driving

    Yingzi Ma, Yulong Cao, Jiachen Sun, Marco Pavone, and Chaowei Xiao. Dolphins: Multimodal language model for driving. In European Conference on Computer Vision, pages 403–420. Springer, 2024. 1

    2024

  20. [28]

    Drama: Joint risk localization and captioning in driving

    Srikanth Malla, Chiho Choi, Isht Dwivedi, Joon Hee Choi, and Jiachen Li. Drama: Joint risk localization and captioning in driving. In Proceedings of the IEEE/CVF winter confer- ence on applications of computer vision , pages 1043–1052,

  21. [29]

    Lingoqa: Visual question answering for au- tonomous driving

    Ana-Maria Marcu, Long Chen, Jan H ¨unermann, Alice Karn- sund, Benoit Hanotte, Prajwal Chidananda, Saurabh Nair, Vijay Badrinarayanan, Alex Kendall, Jamie Shotton, and Oleg Sinavski. Lingoqa: Visual question answering for au- tonomous driving. arXiv preprint arXiv:2312.14115, ...

    2023 arXiv

  22. [30]

    Gpt-4 technical report, 2024

    OpenAI. Gpt-4 technical report, 2024. 3, 6

    2024

  23. [31]

    Idd-x: A multi-view dataset for ego-relative important object localization and explanation in dense and unstructured traffic

    Chirag Parikh, Rohit Saluja, CV Jawahar, and Ravi Ki- ran Sarvadevabhatla. Idd-x: A multi-view dataset for ego-relative important object localization and explanation in dense and unstructured traffic. In 2024 IEEE Inter- national Conference on Robotics and Automation (ICRA) , ...

    2024

  24. [32]

    Traffic-domain video question answering with auto- matic captioning

    Ehsan Qasemi, Jonathan M Francis, and Alessandro Oltra- mari. Traffic-domain video question answering with auto- matic captioning. arXiv preprint arXiv:2307.09636 , 2023. 2

    2023 arXiv

  25. [33]

    Nuscenes-qa: A multi-modal visual ques- tion answering benchmark for autonomous driving scenario

    Tianwen Qian, Jingjing Chen, Linhai Zhuo, Yang Jiao, and Yu-Gang Jiang. Nuscenes-qa: A multi-modal visual ques- tion answering benchmark for autonomous driving scenario. arXiv preprint arXiv:2305.14836, 2023. 1, 2

    2023 arXiv

  26. [34]

    Comet: A neural framework for mt evaluation, 2020

    Ricardo Rei, Craig Stewart, Ana C Farinha, and Alon Lavie. Comet: A neural framework for mt evaluation, 2020. 6, 12

    2020

  27. [35]

    Rank2tell: A multimodal driving dataset for joint importance ranking and reasoning

    Enna Sachdeva, Nakul Agarwal, Suhas Chundi, Sean Roelofs, Jiachen Li, Mykel Kochenderfer, Chiho Choi, and Behzad Dariush. Rank2tell: A multimodal driving dataset for joint importance ranking and reasoning. In Proceedings of the IEEE/CVF winter conference on applications of com...

    2024

  28. [36]

    Mobile-videogpt: Fast and accurate video understanding language model

    Abdelrahman Shaker, Muhammad Maaz, Hamid Rezatofighi, Salman Khan, and Fahad Shahbaz Khan. Mobile-videogpt: Fast and accurate video understanding language model. arxiv, 2025. 3, 6, 7, 15

    2025

  29. [37]

    Video-xl: Extra-long vision language model for hour-scale video understanding

    Yan Shu, Peitian Zhang, Zheng Liu, Minghao Qin, Junjie Zhou, Tiejun Huang, and Bo Zhao. Video-xl: Extra-long vision language model for hour-scale video understanding. arXiv preprint arXiv:2409.14485, 2024. 3, 6, 7, 15

    2024 arXiv

  30. [38]

    Drivelm: Driving with graph visual ques- tion answering

    Chonghao Sima, Katrin Renz, Kashyap Chitta, Li Chen, Hanxue Zhang, Chengen Xie, Ping Luo, Andreas Geiger, and Hongyang Li. Drivelm: Driving with graph visual ques- tion answering. arXiv preprint arXiv:2312.14150, 2023. 2

    2023 arXiv

  31. [39]

    Gemini: A family of highly capable multi- modal models, 2025

    Gemini Team. Gemini: A family of highly capable multi- modal models, 2025. 3, 6, 7

    2025

  32. [40]

    Qwen2.5-vl, 2025

    Qwen Team. Qwen2.5-vl, 2025. 3, 6, 7, 15

    2025

  33. [41]

    Temporal stability of factors af- fecting injury severity in rear-end and non-rear-end crashes: A random parameter approach with heterogeneity in means and variances

    Chenzhu Wang, Fei Chen, Yunlong Zhang, Shuyi Wang, Bin Yu, and Jianchuan Cheng. Temporal stability of factors af- fecting injury severity in rear-end and non-rear-end crashes: A random parameter approach with heterogeneity in means and variances. Analytic Methods in Accident R...

    2022

  34. [42]

    Effects of speed difference on injury severity of freeway rear-end crashes: Insights from correlated joint random parameters bivariate probit models and temporal instability

    Chenzhu Wang, Mohamed Abdel-Aty, and Lei Han. Effects of speed difference on injury severity of freeway rear-end crashes: Insights from correlated joint random parameters bivariate probit models and temporal instability. Analytic Methods in Accident Research, 42:100320, 2024

    2024

  35. [43]

    Chenzhu Wang, Mohamed Abdel-Aty, Lei Han, and Said M. Easa. Analyzing speed-difference impact on freeway joint injury severities of leading-following vehicles using statisti- cal and data-driven models. Accident Analysis & Prevention, 206:107695, 2024

    2024

  36. [44]

    Tunnel crash severity and congestion duration joint evaluation based on cross-stitch networks

    Chenzhu Wang, Mohamed Abdel-Aty, and Lei Han. Tunnel crash severity and congestion duration joint evaluation based on cross-stitch networks. Accident Analysis & Prevention , 213:107942, 2025. 4

    2025

  37. [45]

    Qwen2-vl: Enhancing vision-language model’s perception of the world at any resolution

    Peng Wang, Shuai Bai, Sinan Tan, Shijie Wang, Zhihao Fan, Jinze Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Yang Fan, Kai Dang, Mengfei Du, Xuancheng Ren, Rui Men, Dayiheng Liu, Chang Zhou, Jingren Zhou, and Jun- yang Lin. Qwen2-vl: Enhancing vision-language model’s ...

    2024 arXiv

  38. [46]

    Deepaccident: A motion and accident prediction benchmark for v2x autonomous driving, 2023

    Tianqi Wang, Sukmin Kim, Wenxuan Ji, Enze Xie, Chongjian Ge, Junsong Chen, Zhenguo Li, and Ping Luo. Deepaccident: A motion and accident prediction benchmark for v2x autonomous driving, 2023. 2, 3

    2023

  39. [47]

    Sutd-trafficqa: A question answering benchmark and an efficient network for video rea- soning over traffic events, 2021

    Li Xu, He Huang, and Jun Liu. Sutd-trafficqa: A question answering benchmark and an efficient network for video rea- soning over traffic events, 2021. 2, 3, 4

    2021

  40. [48]

    Explainable object-induced action decision for autonomous vehicles

    Yiran Xu, Xiaoyin Yang, Lihang Gong, Hsuan-Chu Lin, Tz- Ying Wu, Yunsheng Li, and Nuno Vasconcelos. Explainable object-induced action decision for autonomous vehicles. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 9523–9532, 2020. 2

    2020

  41. [49]

    Crandall, and Ella M

    Yu Yao, Mingze Xu, Yuchen Wang, David J. Crandall, and Ella M. Atkins. Unsupervised traffic accident detection in first-person videos. In 2019 IEEE/RSJ International Confer- ence on Intelligent Robots and Systems (IROS) , pages 273– 280, 2019. 2, 3

    2019

  42. [50]

    Crandall

    Yu Yao, Xizi Wang, Mingze Xu, Zelin Pu, Yuchen Wang, Ella Atkins, and David J. Crandall. Dota: Unsupervised de- tection of traffic anomaly in driving videos. IEEE Transac- tions on Pattern Analysis and Machine Intelligence , 45(1): 444–459, 2023. 2, 3, 4

    2023

  43. [51]

    same seman- tics, different structure

    Chen Yin and Zixuan Zhang. A study of sentence similar- ity based on the all-minilm-l6-v2 model with “same seman- tics, different structure” after fine tuning. In Proceedings of the 2024 2nd International Conference on Image, Algo- rithms and Artificial Intelligence (ICIAAI 20...

    2024

  44. [52]

    Ferret: Refer and ground anything anywhere at any granularity

    Haoxuan You, Haotian Zhang, Zhe Gan, Xianzhi Du, Bowen Zhang, Zirui Wang, Liangliang Cao, Shih-Fu Chang, and Yinfei Yang. Ferret: Refer and ground anything anywhere at any granularity. arXiv preprint arXiv:2310.07704, 2023. 1

    2023 arXiv

  45. [53]

    Traffic Accident Bench- mark for Causality Recognition

    Tackgeun You and Bohyung Han. Traffic Accident Bench- mark for Causality Recognition. In ECCV, 2020. 1, 2, 3, 4

    2020

  46. [54]

    Video instruction tuning with synthetic data, 2024

    Yuanhan Zhang, Jinming Wu, Wei Li, Bo Li, Zejun Ma, Zi- wei Liu, and Chunyuan Li. Video instruction tuning with synthetic data, 2024. 3, 6, 7, 15

    2024

  47. [55]

    Xingcheng Zhou, Konstantinos Larintzakis, Hao Guo, Wal- ter Zimmer, Mingyu Liu, Hu Cao, Jiajie Zhang, Venkat- narayanan Lakshminarasimhan, Leah Strand, and Alois C. Knoll. TUMTraffic-VideoQA: A benchmark for uni- fied spatio-temporal video understanding in traffic scenes. IEEE...

    2025

  48. [56]

    Internvl3: Exploring advanced training and test-time recipes for open-source multimodal models, 2025

    Jinguo Zhu, Weiyun Wang, Zhe Chen, Zhaoyang Liu, Shen- glong Ye, Lixin Gu, Hao Tian, Yuchen Duan, Weijie Su, Jie Shao, Zhangwei Gao, Erfei Cui, Xuehui Wang, Yue Cao, Yangzhou Liu, Xingguang Wei, Hongjie Zhang, Haomin Wang, Weiye Xu, Hao Li, Jiahao Wang, Nianchen Deng, Songze L...

    2025

  49. [57]

    sunny day

    Walter Zimmer, Ross Greer, Xingcheng Zhou, Rui Song, Hu Cao, Daniel Lehmberg, Marc Pavel, Ahmed Alaaeldin Ghita, Akshay Gopalkrishnan, Holger Caesar, Mohan M. Trivedi, and Alois C. Knoll. Towards vision zero: The accid3nd dataset. page 10, 2025. 2 Supplementary: VRU-Accident: ...

    2025

  50. [684]

    Atlantis Press, 2024. 15

    2024

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.