pith. sign in

arxiv: 2606.09142 · v1 · pith:YBGVHAVTnew · submitted 2026-06-08 · 💻 cs.CV · cs.AI

Decoding Pedestrian Crossing Intention from Egocentric Vision via Vision Language Models

Danya Li , Xiang Su , Yan Feng , Rico Krueger This is my paper

Pith reviewed 2026-06-27 17:11 UTC · model grok-4.3

classification 💻 cs.CV cs.AI
keywords egocentric visionpedestrian crossing intentionvision language modelsvisual question answeringparameter-efficient fine-tuningeye gazetraffic safety prediction
0
0 comments X

The pith

Vision language models fine-tuned on egocentric video decode pedestrian crossing intent with 14.5 percent higher accuracy than transformer baselines.

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

The paper formulates the task of predicting whether a pedestrian will cross the street as a yes-no visual question answering problem and tests vision language models on short first-person video clips. Zero-shot versions of the models beat random guessing by a modest margin but lack strong traffic reasoning. Parameter-efficient fine-tuning lifts performance, and adding cues such as eye gaze and ego motion produces a further gain that reaches 14.5 percent above a specialized transformer baseline. If the gains hold, the approach supplies a practical route for vehicles to anticipate pedestrian actions from the driver’s viewpoint.

Core claim

The authors establish that fine-tuned vision language models, in particular the Qwen3-VL-2B model supplied with eye gaze and ego-motion signals, achieve a 14.5 percent accuracy improvement over a transformer baseline on egocentric pedestrian intent prediction and thereby set a new state of the art for the task.

What carries the argument

Closed-ended visual question answering formulation of pedestrian intent combined with parameter-efficient fine-tuning of vision language models and the injection of contextual signals such as eye gaze and ego motion.

If this is right

  • Zero-shot vision language models already exceed random guessing but fall short on higher-level traffic reasoning.
  • Parameter-efficient fine-tuning delivers substantial accuracy lifts over both zero-shot and transformer baselines.
  • Incorporating eye gaze, ego motion, and vehicle motion produces additional measurable gains on the intent prediction task.
  • The best reported configuration establishes a new benchmark for egocentric pedestrian crossing intention decoding.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same fine-tuning recipe could be tested on other first-person prediction problems such as driver intent or cyclist behavior.
  • If the models truly learn intent rather than dataset patterns, they should retain accuracy when camera height, lighting, or cultural crossing norms change.
  • Integration into real-time vehicle systems would require checking whether the added latency of the vision language model remains acceptable.

Load-bearing premise

The reported accuracy gains reflect genuine improvement in capturing pedestrian intent rather than dataset-specific fitting or artifacts of how the question is worded.

What would settle it

Running the same fine-tuned Qwen3-VL-2B model on an independent egocentric video dataset collected under different traffic conditions and finding that the 14.5 percent margin over the transformer baseline vanishes or reverses.

Figures

Figures reproduced from arXiv: 2606.09142 by Danya Li, Rico Krueger, Xiang Su, Yan Feng.

Figure 1
Figure 1. Figure 1: A pipeline of our approach. general-purpose egocentric assistants [5], [6], [7]. Extending these capabilities to dynamic urban environments remains an important but underexplored challenge. Addressing this gap is central to proactive AI systems for pedestrian assistance, including navigation support [3], [8], [9] and specialized aids for visually impaired users [10]. Existing research on egocentric predict… view at source ↗
Figure 2
Figure 2. Figure 2: Examples of visual prompt overlay on raw videos. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Zero-shot ablations on temporal input representation. [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Comparison of parameter-efficient fine-tuning strate [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative analysis of Qwen outputs under CoT prompting. We contrast egocentric-only input (left) with gaze [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Effect of context sampling rate under zero-shot [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Validation vs. test performance of fine-tuned Qwen3- [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
read the original abstract

Egocentric vision offers a first-person view of human perception and decision making, yet its potential for traffic-safety prediction remains underexplored. In this work, we study the decoding of pedestrian crossing intentions from short egocentric video clips. We approach this by formulating the task as a closed-ended visual question answering (VQA) problem and leveraging vision language models (VLMs) to predict the pedestrians' intent. We first benchmark three families of state-of-the-art VLMs in a zero-shot setting, finding that they achieve moderate gains over random guessing but exhibit limited higher-level traffic reasoning. Motivated by these findings, we further adapt VLMs to the target task using parameter-efficient fine-tuning. Our results show that the fine-tuned models substantially outperform their zero-shot counterparts and achieve a 9\% accuracy improvement over a specialized transformer-based baseline. Finally, we demonstrate that incorporating additional contextual cues, including ego motion, vehicle motion, and eye gaze, further improves predictive performance. In particular, the fine-tuned Qwen3-VL-2B model guided by eye gaze and ego motion achieves a 14.5% accuracy improvement over the transformer baseline, establishing a new state of the art for egocentric pedestrian intent decoding.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 2 minor

Summary. The paper claims that pedestrian crossing intention decoding from short egocentric video clips can be effectively addressed by reformulating the task as closed-ended visual question answering and leveraging vision-language models. Zero-shot VLMs achieve moderate gains over random guessing but show limited traffic reasoning; parameter-efficient fine-tuning yields a 9% accuracy improvement over a specialized transformer baseline; and adding contextual cues (ego motion, vehicle motion, eye gaze) further boosts performance, with the fine-tuned Qwen3-VL-2B model guided by eye gaze and ego motion delivering a 14.5% accuracy gain over the baseline and establishing a new state of the art.

Significance. If the reported gains are robustly validated, the work would be significant for showing how VLMs can be adapted via parameter-efficient fine-tuning and contextual cues for egocentric traffic-safety prediction, a domain where such models have been underexplored. The empirical benchmarking across VLM families and the demonstration that eye gaze and ego motion provide additive value are concrete contributions that could inform future multimodal intent-prediction systems.

major comments (1)
  1. [Abstract] Abstract: the central SOTA claim rests on stated accuracy improvements of 9% and 14.5% over the transformer baseline, yet the abstract (and by extension the results section) supplies no information on dataset size, number of test instances, statistical significance tests, error bars, or controls for data leakage; these omissions are load-bearing because they prevent assessment of whether the gains reflect genuine modeling advances rather than dataset-specific artifacts.
minor comments (2)
  1. [Methods] The exact wording of the closed-ended VQA questions and the precise integration mechanism for eye-gaze and ego-motion cues into the VLM input should be stated explicitly (likely in §3 or §4) to allow reproduction and to rule out prompt-engineering artifacts.
  2. [Experiments] Figure and table captions would benefit from including the exact evaluation metric (top-1 accuracy?) and the number of runs or seeds used for the reported percentages.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the positive assessment and recommendation of minor revision. We address the single major comment below regarding missing statistical and dataset details.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central SOTA claim rests on stated accuracy improvements of 9% and 14.5% over the transformer baseline, yet the abstract (and by extension the results section) supplies no information on dataset size, number of test instances, statistical significance tests, error bars, or controls for data leakage; these omissions are load-bearing because they prevent assessment of whether the gains reflect genuine modeling advances rather than dataset-specific artifacts.

    Authors: We agree that the abstract and results section should explicitly report dataset size, number of test instances, error bars, statistical significance, and data-leakage controls to allow proper evaluation of the claimed gains. In the revised version we will (1) add the dataset size and test-set cardinality to the abstract, (2) include error bars and the results of significance tests in the results section and tables, and (3) expand the experimental-setup subsection to detail the train/test split procedure and any leakage-prevention measures. These changes directly address the concern without altering the core claims. revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper presents an empirical machine-learning study: zero-shot benchmarking of VLMs on an egocentric VQA task for pedestrian intent, followed by parameter-efficient fine-tuning and ablation of auxiliary inputs (eye gaze, ego motion). All reported gains are measured against an external transformer baseline on held-out data. No equations, self-definitional loops, fitted parameters renamed as predictions, or load-bearing self-citations appear in the abstract or described methodology. The central claim (accuracy lift from fine-tuning plus context) is therefore an experimental outcome, not a quantity forced by construction from the inputs themselves.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

No free parameters, axioms, or invented entities explicitly introduced beyond standard ML practices; the paper relies on existing VLM architectures and fine-tuning methods.

pith-pipeline@v0.9.1-grok · 5748 in / 1212 out tokens · 21062 ms · 2026-06-27T17:11:23.447232+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

53 extracted references · 13 canonical work pages · 7 internal anchors

  1. [1]

    Pedestrian Behavior Prediction Using Deep Learning Methods for Urban Scenarios: A Review,

    C. Zhang and C. Berger, “Pedestrian Behavior Prediction Using Deep Learning Methods for Urban Scenarios: A Review,”IEEE Transactions on Intelligent Transportation Systems, pp. 1–23, 2023

    2023

  2. [2]

    Predicting Pedestrian Crossing Intention in Autonomous Vehicles: A Review,

    F.-G. Landry and M. A. Akhloufi, “Predicting Pedestrian Crossing Intention in Autonomous Vehicles: A Review,”Neurocomputing, vol. 618, p. 129105, Feb. 2025

    2025

  3. [3]

    EgoNav: Egocentric Scene- aware Human Trajectory Prediction,

    W. Wang, C. K. Liu, and M. Kennedy, “EgoNav: Egocentric Scene- aware Human Trajectory Prediction,” Aug. 2024, arXiv:2403.19026 [cs]

    work page arXiv 2024

  4. [4]

    Bridging Perspectives: A Survey on Cross-view Collaborative Intelligence with Egocentric-Exocentric Vision,

    Y . He, Y . Huang, G. Chen, L. Lu, B. Pei, J. Xu, T. Lu, and Y . Sato, “Bridging Perspectives: A Survey on Cross-view Collaborative Intelligence with Egocentric-Exocentric Vision,”International Journal of Computer Vision, vol. 134, no. 2, p. 62, Jan. 2026

    2026

  5. [5]

    Ego4d: Around the world in 3,000 hours of egocentric video,

    K. Grauman, A. Westbury, E. Byrne, Z. Chavis, A. Furnari, R. Girdhar, J. Hamburger, H. Jiang, M. Liu, X. Liuet al., “Ego4d: Around the world in 3,000 hours of egocentric video,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 18 995–19 012

    2022

  6. [6]

    An outlook into the future of egocentric vision,

    C. Plizzari, G. Goletto, A. Furnari, S. Bansal, F. Ragusa, G. M. Farinella, D. Damen, and T. Tommasi, “An outlook into the future of egocentric vision,”International Journal of Computer Vision, vol. 132, no. 11, pp. 4880–4936, 2024

    2024

  7. [7]

    Egolife: Towards egocentric life assistant,

    J. Yang, S. Liu, H. Guo, Y . Dong, X. Zhang, S. Zhang, P. Wang, Z. Zhou, B. Xie, Z. Wanget al., “Egolife: Towards egocentric life assistant,” inProceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp. 28 885–28 900

    2025

  8. [8]

    EgoCogNav: Cognition-aware Human Egocentric Navigation

    Z. Qiu, Z. Liu, W. Niu, T. Bhattacharjee, and S. Kalantari, “EgoCog- Nav: Cognition-aware Human Egocentric Navigation,” Nov. 2025, arXiv:2511.17581 [cs]. 8

    work page internal anchor Pith review arXiv 2025

  9. [9]

    Lookout: Real-world humanoid egocentric navigation,

    B. Pan, A. W. Harley, F. Engelmann, C. K. Liu, and L. J. Guibas, “Lookout: Real-world humanoid egocentric navigation,” inProceed- ings of the IEEE/CVF International Conference on Computer Vision, 2025, pp. 24 977–24 988

    2025

  10. [10]

    Heads-up: Head- mounted egocentric dataset for trajectory prediction in blind assistance systems,

    Y . Haghighi, C. Demonsant, P. Chalimourdas, M. T. Naeini, J. K. Munoz, B. Bacca, S. Suter, M. Gani, and A. Alahi, “Heads-up: Head- mounted egocentric dataset for trajectory prediction in blind assistance systems,”arXiv preprint arXiv:2409.20324, 2024

    work page arXiv 2024

  11. [11]

    Egocentric human trajectory forecasting with a wearable camera and multi-modal fusion,

    J. Qiu, L. Chen, X. Gu, F. P.-W. Lo, Y .-Y . Tsai, J. Sun, J. Liu, and B. Lo, “Egocentric human trajectory forecasting with a wearable camera and multi-modal fusion,”IEEE Robotics and Automation Letters, vol. 7, no. 4, pp. 8799–8806, 2022

    2022

  12. [12]

    KrishnaCam: Using a longitudinal, single-person, egocentric dataset for scene understanding tasks,

    K. K. Singh, K. Fatahalian, and A. A. Efros, “KrishnaCam: Using a longitudinal, single-person, egocentric dataset for scene understanding tasks,” in2016 IEEE Winter Conference on Applications of Computer Vision (WACV), Mar. 2016, pp. 1–9

    2016

  13. [13]

    Egocentric future localization,

    H. S. Park, J.-J. Hwang, Y . Niu, and J. Shi, “Egocentric future localization,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016, pp. 4697–4705

    2016

  14. [14]

    Pedestrian intention prediction for autonomous vehicles: A comprehensive survey,

    N. Sharma, C. Dhiman, and S. Indu, “Pedestrian intention prediction for autonomous vehicles: A comprehensive survey,”Neurocomputing, vol. 508, pp. 120–152, 2022

    2022

  15. [15]

    GPT-4V Takes the Wheel: Promises and Challenges for Pedestrian Behavior Prediction,

    J. Huang, P. Jiang, A. Gautam, and S. Saripalli, “GPT-4V Takes the Wheel: Promises and Challenges for Pedestrian Behavior Prediction,” Proceedings of the AAAI Symposium Series, vol. 3, no. 1, pp. 134–142, May 2024

    2024

  16. [16]

    OmniPredict: GPT-4o Enhanced Multi-modal Pedestrian Crossing Intention Prediction,

    J.-S. Ham, J. Huang, P. Jiang, J. Moon, Y . Kwon, S. Saripalli, and C. Kim, “OmniPredict: GPT-4o Enhanced Multi-modal Pedestrian Crossing Intention Prediction,” Nov. 2024

    2024

  17. [17]

    Seeing beyond frames: Zero-shot pedestrian intention prediction with raw temporal video and multimodal cues,

    P. Zambare, V . N. Thanikella, and Y . Liu, “Seeing beyond frames: Zero-shot pedestrian intention prediction with raw temporal video and multimodal cues,” in2025 3rd International Conference on Artificial Intelligence, Blockchain, and Internet of Things (AIBThings). IEEE, 2025, pp. 1–5

    2025

  18. [18]

    Pedestrian Intention Prediction via Vision-Language Foundation Models,

    M. Azarmi, M. Rezaei, and H. Wang, “Pedestrian Intention Prediction via Vision-Language Foundation Models,” Jul. 2025

    2025

  19. [19]

    Optimizing Vision-Language Model for Road Crossing Intention Estimation,

    R. Uziel and O. Bialer, “Optimizing Vision-Language Model for Road Crossing Intention Estimation,” in2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), Feb. 2025, pp. 1702– 1712

    2025

  20. [20]

    Pedestrian Vision Language Model for Intentions Prediction,

    F. Munir, S. Azam, T. Mihaylova, V . Kyrki, and T. P. Kucner, “Pedestrian Vision Language Model for Intentions Prediction,”IEEE Open Journal of Intelligent Transportation Systems, vol. 6, pp. 393– 406, 2025

    2025

  21. [21]

    Application of Vision-Language Model to Pedestrians Behavior and Scene Under- standing in Autonomous Driving,

    H. Gao, L. Zhang, Y . Zhao, Z. Yang, and J. Cao, “Application of Vision-Language Model to Pedestrians Behavior and Scene Under- standing in Autonomous Driving,” Jul. 2025, arXiv:2501.06680 [cs]

    work page arXiv 2025

  22. [22]

    Vlmped-cot: A large vision-language model with chain-of-thought mechanism for pedestrian crossing intention prediction,

    Y . Ling, Z. Qin, L. Wang, Z. Liu, Y . Liu, and Z. Ma, “Vlmped-cot: A large vision-language model with chain-of-thought mechanism for pedestrian crossing intention prediction,”Communications in Trans- portation Research, 2026

    2026

  23. [23]

    Scaling Egocentric Vision: The EPIC-KITCHENS Dataset

    D. Damen, H. Doughty, G. M. Farinella, S. Fidler, A. Furnari, E. Kazakos, D. Moltisanti, J. Munro, T. Perrett, W. Price, and M. Wray, “Scaling Egocentric Vision: The EPIC-KITCHENS Dataset,” Jul. 2018, arXiv:1804.02748 [cs]

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  24. [24]

    Actor and observer: Joint modeling of first and third-person videos,

    G. A. Sigurdsson, A. Gupta, C. Schmid, A. Farhadi, and K. Alahari, “Actor and observer: Joint modeling of first and third-person videos,” inproceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 7396–7404

    2018

  25. [25]

    Egovlpv2: Egocentric video-language pre-training with fusion in the backbone,

    S. Pramanick, Y . Song, S. Nag, K. Q. Lin, H. Shah, M. Z. Shou, R. Chellappa, and P. Zhang, “Egovlpv2: Egocentric video-language pre-training with fusion in the backbone,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 5285–5297

    2023

  26. [26]

    Egovideo: Exploring egocentric foundation model and downstream adaptation,

    B. Pei, G. Chen, J. Xu, Y . He, Y . Liu, K. Pan, Y . Huang, Y . Wang, T. Lu, L. Wanget al., “Egovideo: Exploring egocentric foundation model and downstream adaptation,”arXiv preprint arXiv:2406.18070, 2024

    work page arXiv 2024

  27. [27]

    Video question answering: Datasets, algorithms and challenges,

    Y . Zhong, W. Ji, J. Xiao, Y . Li, W. Deng, and T.-S. Chua, “Video question answering: Datasets, algorithms and challenges,” inProceed- ings of the 2022 conference on empirical methods in natural language processing, 2022, pp. 6439–6455

    2022

  28. [28]

    Video question answering via gradually refined attention over ap- pearance and motion,

    D. Xu, Z. Zhao, J. Xiao, F. Wu, H. Zhang, X. He, and Y . Zhuang, “Video question answering via gradually refined attention over ap- pearance and motion,” inProceedings of the 25th ACM international conference on Multimedia, 2017, pp. 1645–1653

    2017

  29. [29]

    Tgif-qa: Toward spatio- temporal reasoning in visual question answering,

    Y . Jang, Y . Song, Y . Yu, Y . Kim, and G. Kim, “Tgif-qa: Toward spatio- temporal reasoning in visual question answering,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 2758–2766

    2017

  30. [30]

    Activitynet-qa: A dataset for understanding complex web videos via question answering,

    Z. Yu, D. Xu, J. Yu, T. Yu, Z. Zhao, Y . Zhuang, and D. Tao, “Activitynet-qa: A dataset for understanding complex web videos via question answering,” inProceedings of the AAAI conference on artificial intelligence, vol. 33, no. 01, 2019, pp. 9127–9134

    2019

  31. [31]

    Next-qa: Next phase of question-answering to explaining temporal actions,

    J. Xiao, X. Shang, A. Yao, and T.-S. Chua, “Next-qa: Next phase of question-answering to explaining temporal actions,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2021, pp. 9777–9786

    2021

  32. [32]

    Agqa: A benchmark for compositional spatio-temporal reasoning,

    M. Grunde-McLaughlin, R. Krishna, and M. Agrawala, “Agqa: A benchmark for compositional spatio-temporal reasoning,” inProceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, pp. 11 287–11 297

    2021

  33. [33]

    From representation to reasoning: To- wards both evidence and commonsense reasoning for video question- answering,

    J. Li, L. Niu, and L. Zhang, “From representation to reasoning: To- wards both evidence and commonsense reasoning for video question- answering,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2022

    2022

  34. [34]

    Egotaskqa: Understanding human tasks in egocentric videos,

    B. Jia, T. Lei, S.-C. Zhu, and S. Huang, “Egotaskqa: Understanding human tasks in egocentric videos,”Advances in Neural Information Processing Systems, vol. 35, pp. 3343–3360, 2022

    2022

  35. [35]

    Intentqa: Context-aware video intent reasoning,

    J. Li, P. Wei, W. Han, and L. Fan, “Intentqa: Context-aware video intent reasoning,” inProceedings of the IEEE/CVF international conference on computer vision, 2023, pp. 11 963–11 974

    2023

  36. [36]

    In the eye of mllm: Benchmarking egocentric video intent understanding with gaze-guided prompting,

    T. Peng, J. Hua, M. Liu, and F. Lu, “In the eye of mllm: Benchmarking egocentric video intent understanding with gaze-guided prompting,” arXiv preprint arXiv:2509.07447, 2025

    work page arXiv 2025

  37. [37]

    Qwen3 Technical Report

    Q. Team, “Qwen3 technical report,” 2025. [Online]. Available: https://arxiv.org/abs/2505.09388

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  38. [38]

    InternVL3: Exploring Advanced Training and Test-Time Recipes for Open-Source Multimodal Models

    J. Zhu, W. Wang, Z. Chen, Z. Liu, S. Ye, L. Gu, H. Tian, Y . Duan, W. Su, J. Shaoet al., “Internvl3: Exploring advanced training and test-time recipes for open-source multimodal models,”arXiv preprint arXiv:2504.10479, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  39. [39]

    Grounded question-answering in long egocentric videos,

    S. Di and W. Xie, “Grounded question-answering in long egocentric videos,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 12 934–12 943

    2024

  40. [40]

    Chain-of-thought prompting elicits reasoning in large language models,

    J. Wei, X. Wang, D. Schuurmans, M. Bosma, F. Xia, E. Chi, Q. V . Le, D. Zhouet al., “Chain-of-thought prompting elicits reasoning in large language models,”Advances in neural information processing systems, vol. 35, pp. 24 824–24 837, 2022

    2022

  41. [41]

    Set-of-Mark Prompting Unleashes Extraordinary Visual Grounding in GPT-4V

    J. Yang, H. Zhang, F. Li, X. Zou, C. Li, and J. Gao, “Set-of-mark prompting unleashes extraordinary visual grounding in gpt-4v,”arXiv preprint arXiv:2310.11441, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  42. [42]

    Fine-grained visual prompting,

    L. Yang, Y . Wang, X. Li, X. Wang, and J. Yang, “Fine-grained visual prompting,”Advances in Neural Information Processing Systems, vol. 36, pp. 24 993–25 006, 2023

    2023

  43. [43]

    Large language models are zero-shot reasoners,

    T. Kojima, S. S. Gu, M. Reid, Y . Matsuo, and Y . Iwasawa, “Large language models are zero-shot reasoners,”Advances in neural infor- mation processing systems, vol. 35, pp. 22 199–22 213, 2022

    2022

  44. [44]

    Grounding dino: Marrying dino with grounded pre-training for open-set object detection,

    S. Liu, Z. Zeng, T. Ren, F. Li, H. Zhang, J. Yang, Q. Jiang, C. Li, J. Yang, H. Suet al., “Grounding dino: Marrying dino with grounded pre-training for open-set object detection,” inEuropean conference on computer vision. Springer, 2024, pp. 38–55

    2024

  45. [45]

    Simple online and realtime tracking,

    A. Bewley, Z. Ge, L. Ott, F. Ramos, and B. Upcroft, “Simple online and realtime tracking,” in2016 IEEE international conference on image processing (ICIP). Ieee, 2016, pp. 3464–3468

    2016

  46. [46]

    Analyzing the behaviors of pedestrians and cyclists in interactions with autonomous systems using controlled experiments: A literature review,

    D. Li, W. Mao, F. C. Pereira, Y . Xiao, X. Su, and R. Krueger, “Analyzing the behaviors of pedestrians and cyclists in interactions with autonomous systems using controlled experiments: A literature review,”Transportation Research Part F: Traffic Psychology and Behaviour, vol. 114, pp. 270–307, 2025

    2025

  47. [48]

    Challenges and trends in egocentric vision: A survey,

    X. Li, H. Qiu, L. Wang, H. Zhang, C. Qi, L. Han, H. Xiong, and H. Li, “Challenges and trends in egocentric vision: A survey,”Machine Intelligence Research, vol. 23, no. 1, pp. 1–33, 2026

    2026

  48. [49]

    Eye Gaze-Informed and Context-Aware Pedestrian Trajectory Prediction in Shared Spaces with Automated Shuttles: A Virtual Reality Study

    D. Li, Y . Feng, and R. Krueger, “Eye Gaze-Informed and Context-Aware Pedestrian Trajectory Prediction in Shared Spaces 9 with Automated Shuttles: A Virtual Reality Study,” Mar. 2026, arXiv:2603.19812 [cs]

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  49. [50]

    Lora: Low-rank adaptation of large language models

    E. J. Hu, Y . Shen, P. Wallis, Z. Allen-Zhu, Y . Li, S. Wang, L. Wang, W. Chenet al., “Lora: Low-rank adaptation of large language models.” Iclr, vol. 1, no. 2, p. 3, 2022

    2022

  50. [51]

    Learning Transferable Visual Models From Natural Language Su- pervision,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clark, G. Krueger, and I. Sutskever, “Learning Transferable Visual Models From Natural Language Su- pervision,” inProceedings of the 38th International Conference on Machine Learning. PMLR, Jul. 2021, pp. 8748–8763

    2021

  51. [52]

    Advancing egocentric video question answering with multimodal large language models.arXiv preprint arXiv:2504.04550, 2025

    A. Patel, V . Chitalia, and Y . Yang, “Advancing Egocentric Video Question Answering with Multimodal Large Language Models,” Apr. 2025, arXiv:2504.04550 [cs]

    work page arXiv 2025

  52. [53]

    LatentOmni: Rethinking Omni-Modal Understanding via Unified Audio-Visual Latent Reasoning

    Y . Dai, Z. Wu, B. Zeng, D. Hua, J. Liu, B. Li, Y . Wang, C. Tong, H. Liang, X. Maet al., “Latentomni: Rethinking omni-modal un- derstanding via unified audio-visual latent reasoning,”arXiv preprint arXiv:2605.22012, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  53. [54]

    Chrono: A simple blueprint for representing time in mllms,

    B. Meinardus, H. G. Rodriguez, A. Batra, A. Rohrbach, and M. Rohrbach, “Chrono: A simple blueprint for representing time in mllms,” inProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops, October 2025, pp. 4151–4156. 10

    2025