Pith. sign in

REVIEW 6 major objections 6 minor 57 references

Pedestrian Intention and Trajectory Prediction in Unstructured Traffic Using IDD-PeD

T0 review · 6 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 Unstructured traffic cuts pedestrian intent accuracy by 15%.

desk verdict Useful new dataset, but the headline numbers and cross-dataset protocol don't yet support the robustness claims. read the letter →

arxiv 2506.22111 v1 pith:EYCV5ZUE submitted 2025-06-27 cs.CV cs.HC

Ruthvik Bokkasam , Shankar Gangisetty , A. H. Abdul Hafez , C. V. Jawahar This is my paper
classification cs.CVcs.HC
keywords pedestrianintentionpredictiontrajectoryunstructuredtrafficautonomousdrivingdatasetbenchmarkocclusionnightlightvehicle-pedestrianinteraction
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 argues that pedestrian behavior prediction models, trained and tested on structured traffic datasets, are not robust in unstructured environments such as dense Indian streets without signals or crosswalks. To make this case, it introduces IDD-PeD, a dataset of over 650,000 pedestrian bounding boxes with 19 behavioral attributes from 100+ hours of driving video. Benchmarking eight intention-prediction and four trajectory-prediction methods, it reports that intention accuracy drops by up to 15% and trajectory error rises by up to 1,208 MSE relative to standard datasets like PIE and JAAD. The paper's purpose is to establish that unstructured traffic poses distinct challenges—occlusion, poor nighttime illumination, unsignalized scenes, and vehicle-pedestrian interactions—that current datasets and models do not capture.

What carries the argument

The load-bearing object is the IDD-PeD dataset, built from 100+ hours of ego-mounted camera footage in dense South Asian urban traffic, with 205,145 annotated frames, 494,854 pedestrian boxes, and 190,395 additional traffic-object boxes. It provides five annotation types—spatial, behavioral, scene, interaction, and location—where behavioral annotations cover six classes of attributes such as crossing behavior, traffic interaction, pedestrian activity, attention indicators, group dynamics, and stationary behavior, totaling 19 behavioral attributes. The dataset's role is to expose distribution shift: the same eight intention baselines and four trajectory baselines run on PIE and JAAD lose accuracy when evaluated on IDD-PeD under the same observation and prediction settings.

What would settle it

Retrain or fine-tune each baseline on IDD-PeD using the original public training protocol for PIE and JAAD, and measure the drop; if the drop shrinks or disappears, the claimed difficulty of unstructured scenes is an artifact of the evaluation setup. A complementary check is to train on IDD-PeD and test on PIE and JAAD: if those numbers stay near structured-scene levels, the dataset is merely different rather than harder.

Watch

Extended reading notes

Core claim

The central claim is that IDD-PeD is the largest pedestrian behavior dataset for unstructured traffic, and that state-of-the-art pedestrian intention and trajectory models underperform on it by large margins compared with their results on PIE and JAAD. The paper attributes the degradation to four factors its dataset explicitly annotates: occlusions, illumination changes, absence of signals and crosswalks, and pedestrian-vehicle interactions. From the experiments, the paper concludes that existing visual-feature-heavy deterministic trajectory models are particularly fragile, while bounding-box-based stochastic models degrade less, and that no current intention model maintains its structured-scene performance in these settings.

Load-bearing premise

The comparison assumes that the baseline methods were run on IDD-PeD in a way that is exactly as fair as on PIE and JAAD, but the paper does not report the training protocol, hyperparameters, or data splits for those baselines, so the performance drop could partly reflect implementation differences rather than intrinsic dataset difficulty.

Editorial extensions

If this is right

  • Autonomous driving systems deployed in unstructured traffic need models that are trained or at least fine-tuned on data like IDD-PeD; relying on PIE/JAAD-trained models is shown to be unsafe by the reported margins.
  • Occlusion and nightlight are the most consistent causes of degradation across intention baselines, so robustness methods for perception under occlusion and low illumination become prerequisites for accurate prediction.
  • Bounding-box-based stochastic trajectory models (SGNet, BiTraP) degrade far less than visual-feature-based deterministic ones, suggesting representations that decouple detection from appearance are a promising axis for improving transfer.
  • The dataset's interaction annotations could support new objectives, such as predicting when a pedestrian will change intent, that are not available in structured datasets.

Reading between the lines

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

  • A direct extension the paper does not run is the reverse transfer experiment—training on IDD-PeD and testing on PIE/JAAD—which would tell whether the dataset is genuinely harder or just draws from a different distribution; the paper's one-directional comparison cannot answer this.
  • The reported 'rolling behavior' and interaction labels could be used to define a new metric, such as intent-switch frequency, that would quantify how often a pedestrian changes their decision—an event type that the current binary crossing/not-crossing label collapses.
  • Since the paper finds a 15% AUC drop for the strongest intention baseline, the dataset may serve as a stress test for domain adaptation and self-supervised pretraining methods, which the paper leaves unexamined.
  • The authors note MaskPCPA's underperformance despite its strong results on PIE/JAAD, but a testable explanation they hint at is parameter capacity; this could be checked by scaling MaskPCPA to the same parameter count as PCPA.
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

6 major / 6 minor

Summary. The paper introduces IDD-PeD, a new pedestrian intention and trajectory prediction dataset collected from Indian unstructured traffic environments. It provides 494,854 pedestrian bounding boxes and 190,395 additional traffic-object boxes across 205,145 annotated frames, with behavioral, scene, interaction, location, and occlusion annotations. The authors benchmark eight pedestrian intention prediction (PIP) baselines and four pedestrian trajectory prediction (PTP) baselines, reporting that intention prediction performance drops by up to 15% and trajectory MSE increases by up to 1208 relative to PIE and JAAD, and they claim that IDD-PeD is the largest dataset of its kind. The paper positions the dataset as a harder test bed for pedestrian behavior modeling in unstructured, dense traffic.

Significance. If the dataset is released and the benchmark numbers are trustworthy, IDD-PeD would be a valuable community resource. It addresses an underexplored regime—unstructured South Asian traffic with nightlight, occlusion, unsignalized crossings, and explicit vehicle-pedestrian interaction annotations—and it evaluates a broad set of standard baselines. The paper correctly identifies weaknesses in existing datasets such as PIE, JAAD, and TITAN with respect to these conditions. However, the quantitative claims that give the paper its headline results are not currently supported: the abstract's 'up to 15%' and 'up to 1208 MSE' statements are internally contradicted by the paper's own tables, the cross-dataset MSE comparisons use unnormalized pixel coordinates across different camera systems, and the baseline protocol is under-specified. The dataset contribution itself is potentially significant, but the experimental evidence needs substantial revision before the robustness conclusion can be accepted.

major comments (6)
  1. [Abstract and Table II] The abstract claims 'a significant performance drop of up to 15%' for intention prediction, but Table II shows MaskPCPA dropping from 0.86 AUC on PIE to 0.61 AUC on IDD-PeD, which is a 0.25 absolute drop, and from 0.80 to 0.25 F1, a 0.55 drop. The stated 'up to 15%' is therefore contradicted by the paper's own results. Please reconcile the abstract and Section IV.B with the full range of drops in Table II.
  2. [Section IV.A and Table III] The central trajectory comparison uses raw MSE in squared pixels across datasets recorded with different cameras and resolutions. PIE is 1920x1080, while IDD-PeD uses GoPro Hero 8 and DDPAI dashcam in unspecified 'HD' modes; Section III.A does not state the output resolutions. For the same physical prediction error, higher-resolution images yield larger pixel MSE, so the reported gaps (e.g., MTN 444 to 1652, PIETraj 559 to 2181) conflate camera properties with dataset difficulty. The paper should either report normalized coordinates, calibrate to a metric space using the OBD/GPS data, or clearly restrict all cross-dataset comparisons to datasets with identical resolution and capture settings.
  3. [Section IV.A] The experimental protocol is insufficiently specified to support the benchmark claims. The paper states only that 'the same observation and time settings as in [31] for PIP and [12] for PTP' were adopted; it does not describe the training/validation split for PIE and JAAD, whether the PIE and JAAD numbers were recomputed or taken from the original papers, the hyperparameter selection process per dataset, input resolutions, or the class balance of crossing vs. non-crossing labels. Without this information, the reported performance drops could reflect protocol differences rather than intrinsic dataset difficulty. Please provide full implementation details and, ideally, release code and trained models.
  4. [Contributions bullet and Table I] The claim that IDD-PeD is 'the largest dataset' is contradicted by the paper's own Table I: PIE has 750K bounding boxes while IDD-PeD has 686K total boxes, and Section III.B states that only 494,854 of the IDD-PeD boxes are pedestrian boxes. The 'more than 650K pedestrian bounding boxes' wording in the contributions and abstract is also inaccurate because the 650K+ figure includes 190,395 non-pedestrian traffic object boxes. Please correct the claim to refer to the largest dataset for unstructured Indian traffic, or compare only pedestrian-specific annotations.
  5. [Abstract and Table III] The abstract's 'increase of up to 1208 MSE' is not the maximum increase shown in Table III. PIETraj has an MSE increase of 1622 (559 to 2181) and a CF-MSE increase of 6798 (2162 to 8960). Even limiting to the deterministic methods mentioned in Section IV.C, the 1208 figure corresponds only to MTN, not to the largest observed gap. Please correct the summary statistic or state explicitly which method and metric are being referenced.
  6. [Section IV.A and Tables IV, V] The paper does not report the number of samples or class priors for the challenge-based splits in Tables IV and V. Low F1 scores on IDD-PeD (e.g., 0.15–0.33 in Table II) could indicate severe label imbalance rather than intrinsic difficulty, and the 'w/o' and 'w' column headers for occlusion, signalized type, and interaction are not defined. Please report per-split sample counts, label distributions, and a precise definition of each condition.
minor comments (6)
  1. [Abstract and Section I] The phrase 'defeating standard pedestrian datasets' is unclear and should be rewritten, e.g., 'in comparison to standard pedestrian datasets'.
  2. [Section V] The word 'higlights' is a typo for 'highlights'.
  3. [Section III.B] The relationship between the claimed '19 behavioral attributes' and the 'six distinct classes' of behavioral annotations is not explained; please clarify the hierarchy or count.
  4. [Section III.C] The text says the dataset is split into 70% training and 30% testing, with no validation set. For reproducibility and model selection, please either provide a validation split or describe how hyperparameters were selected.
  5. [References] Reference [38] appears to have a malformed title ('Trep: Transformer-based evidential prediction for pedestrian intention with uncertainty' is followed by an incomplete phrase), and several references lack page numbers or venue formatting; please check the bibliography.
  6. [Figure 3 and Table I] The figure captions inconsistently refer to 'signalized types' and 'crosswalk (C), signal (S), crosswalk and signal (CS), and absence of crosswalk and signal (NA)' without defining the abbreviations in one place; please unify the notation.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the dataset and benchmark claims rest on external model evaluations and dataset statistics, not on fit parameters or self-citation chains.

full rationale

The paper's central claims are empirical: it introduces IDD-PeD and reports how existing intention and trajectory prediction models perform when evaluated on it. No parameter is fitted to a subset of the data and then renamed a prediction, and no stated result is equivalent by construction to an input. The cross-dataset MSE comparisons are indeed vulnerable to protocol and camera-resolution confounds, and the '650K pedestrian bounding boxes' claim conflicts with the paper's own stated 494,854 pedestrian boxes, but those are correctness risks, not circularity. The self-citations (IDD-3D, IDD-X) are used only to motivate unstructured traffic as a research context; they are not load-bearing for the benchmark logic. No uniqueness theorem, ansatz, or definitional equivalence is imported. Therefore the derivation chain is self-contained with respect to circularity, and the appropriate score is 0.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

The paper introduces no free parameters or invented entities; the central claims rest on dataset construction and benchmarking choices. The main assumptions are the evaluation protocol and the split, which are standard but under-detailed.

assumptions (3)
  • domain assumption The chosen prediction horizon (1.5s for trajectory, 1-2s time-to-event for intention) and observation window (0.5s) are representative of real driving needs.
    Section IV.A sets these values but provides no justification that they are the most safety-relevant or comparable across datasets.
  • domain assumption The 70/30 train/test split of IDD-PeD yields statistically meaningful evaluation.
    Section III.C states the split without showing class balance across splits or sequence-level separation to prevent leakage.
  • domain assumption The baseline implementations and hyperparameters used here match the original publications' intent.
    Section IV.A says 'fair comparison' but gives no training recipe, so results could differ from the original papers.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Pedestrian Intention and Trajectory Prediction in Unstructured Traffic Using IDD-PeD." pith.science (2026). https://pith.science/paper/EYCV5ZUE

@misc{pith2026250622111,
  author       = {Pith},
  title        = {Pith review of: Pedestrian Intention and Trajectory Prediction in Unstructured Traffic Using IDD-PeD},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EYCV5ZUE}},
  note         = {Machine review of arXiv:2506.22111}
}
abstract

With the rapid advancements in autonomous driving, accurately predicting pedestrian behavior has become essential for ensuring safety in complex and unpredictable traffic conditions. The growing interest in this challenge highlights the need for comprehensive datasets that capture unstructured environments, enabling the development of more robust prediction models to enhance pedestrian safety and vehicle navigation. In this paper, we introduce an Indian driving pedestrian dataset designed to address the complexities of modeling pedestrian behavior in unstructured environments, such as illumination changes, occlusion of pedestrians, unsignalized scene types and vehicle-pedestrian interactions. The dataset provides high-level and detailed low-level comprehensive annotations focused on pedestrians requiring the ego-vehicle's attention. Evaluation of the state-of-the-art intention prediction methods on our dataset shows a significant performance drop of up to $\mathbf{15\%}$, while trajectory prediction methods underperform with an increase of up to $\mathbf{1208}$ MSE, defeating standard pedestrian datasets. Additionally, we present exhaustive quantitative and qualitative analysis of intention and trajectory baselines. We believe that our dataset will open new challenges for the pedestrian behavior research community to build robust models. Project Page: https://cvit.iiit.ac.in/research/projects/cvit-projects/iddped

Figures

Figures reproduced from arXiv: 2506.22111 by the authors.

Figure 1
Figure 1. Illustration of pedestrian intention and trajectory encountering various challenges within our unstructured traffic IDD-PeD dataset. The challenges include occlusions, signalized types, vehicle-pedestrian interactions, and illumination changes. A comparative analysis of these challenges with other datasets is presented in Table I. Intent of C:Crossing with trajectory and NC:Not Crossing. Abstract— With the rapid adv… view at source ↗
Figure 2
Figure 2. Rolling behavior is common in unstructured environments, where pedestrians hesitate, pause, or change direction unpredictably, even after they start to cross. (left) In our dataset, due to rolling behaviour, the pedestrian does not cross and the ego-vehicle moves ahead, while (right) in PIE, the pedestrian crosses since structured settings typically prompt the ego-vehicle to yield giving priority to the pedestrian. … view at source ↗
Figure 3
Figure 3. Annotation instances and data statistics of IDD-PeD. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Qualitative evaluation of the best and worst PTP models [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

57 extracted references · 48 canonical work pages

  1. [31]

    Benchmark for evaluating pedestrian action prediction,

    I. Kotseruba, A. Rasouli, and J. K. Tsotsos, “Benchmark for evaluating pedestrian action prediction,” in Proceedings of the IEEE/CVF winter conference on applications of computer vision , pp. 1258–1268, 2021

    work page 2021

  2. [12]

    Pie: A large-scale dataset and models for pedestrian intention estimation and trajectory prediction,

    A. Rasouli, I. Kotseruba, T. Kunic, and J. K. Tsotsos, “Pie: A large-scale dataset and models for pedestrian intention estimation and trajectory prediction,” in Proceedings of the IEEE/CVF International Conference on Computer Vision , pp. 6262–6271, 2019

    work page 2019

  3. [1]

    Intent-aware pedestrian prediction for adaptive crowd navigation,

    K. D. Katyal, G. D. Hager, and C.-M. Huang, “Intent-aware pedestrian prediction for adaptive crowd navigation,” in 2020 IEEE International Conference on Robotics and Automation (ICRA) , pp. 3277–3283, 2020

    work page 2020

  4. [2]

    Intention aware robot crowd navigation with attention-based interaction graph,

    S. Liu, P. Chang, Z. Huang, N. Chakraborty, K. Hong, W. Liang, D. L. McPherson, J. Geng, and K. Driggs-Campbell, “Intention aware robot crowd navigation with attention-based interaction graph,” in 2023 IEEE International Conference on Robotics and Automation (ICRA) , pp. 12015–12021, 2023

    work page 2023

  5. [3]

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

    C. Parikh, R. Saluja, C. V . Jawahar, and R. Kiran Sarvadevabhatla, “Idd-x: A multi-view dataset for ego-relative important object local- ization and explanation in dense and unstructured traffic,” in 2024 IEEE International Conference on Robotics and Automation (ICRA) , pp. 14815–14821, 2024

    work page 2024

  6. [4]

    Argoverse 2: Next generation datasets for self-driving perception and forecasting,

    B. Wilson, W. Qi, T. Agarwal, J. Lambert, J. Singh, S. Khandelwal, B. Pan, R. Kumar, A. Hartnett, J. K. Pontes, D. Ramanan, P. Carr, and J. Hays, “Argoverse 2: Next generation datasets for self-driving perception and forecasting,” in NeurIPS Datasets and Benchmarks , 2021

    work page 2021

  7. [5]

    nuscenes: A multimodal dataset for autonomous driving,

    H. Caesar, V . Bankiti, A. H. Lang, S. V ora, V . E. Liong, Q. Xu, A. Krishnan, Y . Pan, G. Baldan, and O. Beijbom, “nuscenes: A multimodal dataset for autonomous driving,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pp. 11621–11631, 2020

    work page 2020

  8. [6]

    Idd-3d: Indian driving dataset for 3d unstructured road scenes,

    S. Dokania, A. H. A. Hafez, A. Subramanian, M. Chandraker, and C. Jawahar, “Idd-3d: Indian driving dataset for 3d unstructured road scenes,” in 2023 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), pp. 4471–4480, 2023

    work page 2023

Show all 57 references
  1. [7]

    Scalability in perception for autonomous driving: Waymo open dataset,

    P. Sun, H. Kretzschmar, X. Dotiwalla, A. Chouard, V . Patnaik, P. Tsui, J. Guo, Y . Zhou, Y . Chai, B. Caine, et al., “Scalability in perception for autonomous driving: Waymo open dataset,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , ...

    2020

  2. [8]

    Crowds by example,

    A. Lerner, Y . Chrysanthou, and D. Lischinski, “Crowds by example,” in Computer graphics forum , vol. 26, pp. 655–664, Wiley Online Library, 2007

    2007

  3. [9]

    You’ll never walk alone: Modeling social behavior for multi-target tracking,

    S. Pellegrini, A. Ess, K. Schindler, and L. V . Gool, “You’ll never walk alone: Modeling social behavior for multi-target tracking,” in ICCV, pp. 261–268, IEEE Computer Society, 2009

    2009

  4. [10]

    Learning social etiquette: Human trajectory understanding in crowded scenes,

    A. Robicquet, A. Sadeghian, A. Alahi, and S. Savarese, “Learning social etiquette: Human trajectory understanding in crowded scenes,” in Computer Vision–ECCV 2016: 14th European Conference, Amster- dam, The Netherlands, October 11-14, 2016, Proceedings, Part VIII 14, pp. 549–5...

    2016

  5. [11]

    Are they going to cross? a benchmark dataset and baseline for pedestrian crosswalk behavior,

    A. Rasouli, I. Kotseruba, and J. K. Tsotsos, “Are they going to cross? a benchmark dataset and baseline for pedestrian crosswalk behavior,” in Proceedings of the IEEE International Conference on Computer Vision Workshops, pp. 206–213, 2017

    2017

  6. [13]

    TITAN: Future forecast using ac- tion priors,

    S. Malla, B. Dariush, and C. Choi, “TITAN: Future forecast using ac- tion priors,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pp. 11186–11196, 2020

    2020

  7. [14]

    Psi: A pedestrian behavior dataset for socially intelligent autonomous car,

    T. Chen, T. Jing, R. Tian, Y . Chen, J. Domeyer, H. Toyoda, R. Sherony, and Z. Ding, “Psi: A pedestrian behavior dataset for socially intelligent autonomous car,” arXiv preprint arXiv:2112.02604 , 2021

    2021 arXiv

  8. [15]

    Euro-pvi: Pedestrian vehicle interactions in dense urban centers,

    A. Bhattacharyya, D. O. Reino, M. Fritz, and B. Schiele, “Euro-pvi: Pedestrian vehicle interactions in dense urban centers,” in Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 6408–6417, 2021

    2021

  9. [16]

    Tbd pedestrian data collection: Towards rich, portable, and large- scale natural pedestrian data,

    A. Wang, D. Sato, Y . Corzo, S. Simkin, A. Biswas, and A. Steinfeld, “Tbd pedestrian data collection: Towards rich, portable, and large- scale natural pedestrian data,” in 2024 IEEE International Conference on Robotics and Automation (ICRA) , pp. 637–644, 2024

    2024

  10. [17]

    Nightowls: A pedestrians at night dataset,

    L. Neumann, M. Karg, S. Zhang, C. Scharfenberger, E. Piegert, S. Mistr, O. Prokofyeva, R. Thiel, A. Vedaldi, A. Zisserman, et al. , “Nightowls: A pedestrians at night dataset,” inComputer Vision–ACCV 2018: 14th Asian Conference on Computer Vision, Perth, Australia, December 2–...

    2018

  11. [18]

    Jaywalkervr: A vr system for collecting safety-critical pedestrian-vehicle interactions,

    K. Mukoya, E. Weng, R. Choudhury, and K. Kitani, “Jaywalkervr: A vr system for collecting safety-critical pedestrian-vehicle interactions,” in 2024 IEEE International Conference on Robotics and Automation (ICRA), pp. 9600–9607, 2024

    2024

  12. [19]

    Pedestrian temporal gap acceptance behavior at unsignalized intersections in kanpur, india,

    V . Vasudevan, M. Mehta, and B. Dutta, “Pedestrian temporal gap acceptance behavior at unsignalized intersections in kanpur, india,” Transportation research part F: traffic psychology and behaviour , vol. 74, pp. 95–103, 2020

    2020

  13. [20]

    Spatiotemporal relationship reasoning for pedestrian intent prediction,

    B. Liu, E. Adeli, Z. Cao, K.-H. Lee, A. Shenoi, A. Gaidon, and J. C. Niebles, “Spatiotemporal relationship reasoning for pedestrian intent prediction,” 2020

    2020

  14. [21]

    Very deep convolutional networks for large-scale image recognition,

    K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image recognition,” arXiv preprint arXiv:1409.1556 , 2014

    2014 arXiv

  15. [22]

    Rnn-based pedestrian crossing prediction using activity and pose-related features,

    J. Lorenzo, I. Parra, F. Wirth, C. Stiller, D. F. Llorca, and M. A. Sotelo, “Rnn-based pedestrian crossing prediction using activity and pose-related features,” in 2020 IEEE Intelligent Vehicles Symposium (IV), pp. 1801–1806, IEEE, 2020

    2020

  16. [23]

    Do they want to cross? understanding pedestrian intention for behavior prediction,

    I. Kotseruba, A. Rasouli, and J. K. Tsotsos, “Do they want to cross? understanding pedestrian intention for behavior prediction,” in 2020 IEEE Intelligent Vehicles Symposium (IV) , pp. 1688–1693, IEEE, 2020

    2020

  17. [24]

    Real-time intent prediction of pedestrians for autonomous ground vehicles via spatio-temporal densenet,

    K. Saleh, M. Hossny, and S. Nahavandi, “Real-time intent prediction of pedestrians for autonomous ground vehicles via spatio-temporal densenet,” in 2019 International Conference on Robotics and Automa- tion (ICRA), pp. 9704–9710, IEEE, 2019

    2019

  18. [25]

    Spatio-temporal densenet for real-time intent prediction of pedestrians in urban traffic environ- ments,

    K. Saleh, M. Hossny, and S. Nahavandi, “Spatio-temporal densenet for real-time intent prediction of pedestrians in urban traffic environ- ments,” Neurocomputing, vol. 386, pp. 317–324, 2020

    2020

  19. [26]

    Learning spatiotemporal features with 3d convolutional networks,

    D. Tran, L. Bourdev, R. Fergus, L. Torresani, and M. Paluri, “Learning spatiotemporal features with 3d convolutional networks,” in Pro- ceedings of the IEEE international conference on computer vision , pp. 4489–4497, 2015

    2015

  20. [27]

    Quo vadis, action recognition? a new model and the kinetics dataset,

    J. Carreira and A. Zisserman, “Quo vadis, action recognition? a new model and the kinetics dataset,” inproceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pp. 6299–6308, 2017

    2017

  21. [28]

    Pedestrian-driver communication and decision strategies at marked crossings,

    M. Sucha, D. Dostal, and R. Risser, “Pedestrian-driver communication and decision strategies at marked crossings,” Accident Analysis & Prevention, vol. 102, pp. 41–50, 2017

    2017

  22. [29]

    Pedestrian graph: Pedestrian crossing prediction based on 2d pose estimation and graph convolutional networks,

    P. R. G. Cadena, M. Yang, Y . Qian, and C. Wang, “Pedestrian graph: Pedestrian crossing prediction based on 2d pose estimation and graph convolutional networks,” in 2019 IEEE Intelligent Transportation Systems Conference (ITSC) , pp. 2000–2005, IEEE, 2019

    2019

  23. [30]

    Pedestrian action antici- pation using contextual feature fusion in stacked rnns,

    A. Rasouli, I. Kotseruba, and J. K. Tsotsos, “Pedestrian action antici- pation using contextual feature fusion in stacked rnns,” arXiv preprint arXiv:2005.06582, 2020

    2005 arXiv

  24. [32]

    Predicting pedestrian crossing intention with feature fusion and spatio-temporal attention,

    D. Yang, H. Zhang, E. Yurtsever, K. A. Redmill, and ¨U. ¨Ozg¨uner, “Predicting pedestrian crossing intention with feature fusion and spatio-temporal attention,” IEEE Transactions on Intelligent Vehicles , vol. 7, no. 2, pp. 221–230, 2022

    2022

  25. [33]

    Multi-modal hybrid architecture for pedestrian action prediction,

    A. Rasouli, T. Yau, M. Rohani, and J. Luo, “Multi-modal hybrid architecture for pedestrian action prediction,” in 2022 IEEE intelligent Vehicles symposium (IV), pp. 91–97, IEEE, 2022

    2022

  26. [34]

    Action-vit: Pedestrian intent prediction in traffic scenes,

    S. Zhao, H. Li, Q. Ke, L. Liu, and R. Zhang, “Action-vit: Pedestrian intent prediction in traffic scenes,” IEEE Signal Processing Letters , vol. 29, pp. 324–328, 2021

    2021

  27. [35]

    Is attention to bounding boxes all you need for pedestrian action predic- tion?,

    L. Achaji, J. Moreau, T. Fouqueray, F. Aioun, and F. Charpillet, “Is attention to bounding boxes all you need for pedestrian action predic- tion?,” in 2022 IEEE Intelligent Vehicles Symposium (IV) , pp. 895– 902, IEEE, 2022

    2022

  28. [36]

    Pit: Progressive interaction transformer for pedestrian crossing intention prediction,

    Y . Zhou, G. Tan, R. Zhong, Y . Li, and C. Gou, “Pit: Progressive interaction transformer for pedestrian crossing intention prediction,” IEEE Transactions on Intelligent Transportation Systems , 2023

    2023

  29. [37]

    Capformer: Pedestrian crossing action prediction using transformer,

    J. Lorenzo, I. P. Alonso, R. Izquierdo, A. L. Ballardini, ´A. H. Saz, D. F. Llorca, and M. ´A. Sotelo, “Capformer: Pedestrian crossing action prediction using transformer,” Sensors, vol. 21, no. 17, p. 5694, 2021

    2021

  30. [38]

    Trep: Transformer-based evidential prediction for pedestrian intention with uncertainty,

    Z. Zhang, R. Tian, and Z. Ding, “Trep: Transformer-based evidential prediction for pedestrian intention with uncertainty,” in Proceedings of the AAAI Conference on Artificial Intelligence, vol. 37, pp. 3534–3542, 2023

    2023

  31. [39]

    Mdst-dgcn: A mul- tilevel dynamic spatiotemporal directed graph convolutional network for pedestrian trajectory prediction,

    S. Liu, H. Liu, Y . Wang, J. Sun, and T. Mao, “Mdst-dgcn: A mul- tilevel dynamic spatiotemporal directed graph convolutional network for pedestrian trajectory prediction,” Computational intelligence and neuroscience, vol. 2022, no. 1, p. 4192367, 2022

    2022

  32. [40]

    Pedestrian trajectory prediction using dynamics-based deep learning,

    H. Wang, W. Zhi, G. Batista, and R. Chandra, “Pedestrian trajectory prediction using dynamics-based deep learning,” in 2024 IEEE Inter- national Conference on Robotics and Automation (ICRA) , pp. 15068– 15075, IEEE, 2024

    2024

  33. [41]

    Pedestrian intention estimation and trajectory prediction based on data and knowledge- driven method,

    J. Zhou, X. Bai, W. Fu, B. Ning, and R. Li, “Pedestrian intention estimation and trajectory prediction based on data and knowledge- driven method,” IET Intelligent Transport Systems , vol. 18, no. 2, pp. 315–331, 2024

    2024

  34. [42]

    Multimodal transformer networks for pedestrian trajectory prediction.,

    Z. Yin, R. Liu, Z. Xiong, and Z. Yuan, “Multimodal transformer networks for pedestrian trajectory prediction.,” in IJCAI, pp. 1259– 1265, 2021

    2021

  35. [43]

    Stochastic sampling simulation for pedestrian trajectory prediction,

    C. Anderson, X. Du, R. Vasudevan, and M. Johnson-Roberson, “Stochastic sampling simulation for pedestrian trajectory prediction,” in 2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 4236–4243, IEEE, 2019

    2019

  36. [44]

    Bitrap: Bi-directional pedestrian trajectory prediction with multi- modal goal estimation,

    Y . Yao, E. Atkins, M. Johnson-Roberson, R. Vasudevan, and X. Du, “Bitrap: Bi-directional pedestrian trajectory prediction with multi- modal goal estimation,” IEEE Robotics and Automation Letters, vol. 6, no. 2, pp. 1463–1470, 2021

    2021

  37. [45]

    A novel benchmarking paradigm and a scale-and motion- aware model for egocentric pedestrian trajectory prediction,

    A. Rasouli, “A novel benchmarking paradigm and a scale-and motion- aware model for egocentric pedestrian trajectory prediction,” in 2024 IEEE International Conference on Robotics and Automation (ICRA) , pp. 5630–5636, IEEE, 2024

    2024

  38. [46]

    Stepwise goal-driven networks for trajectory prediction,

    C. Wang, Y . Wang, M. Xu, and D. J. Crandall, “Stepwise goal-driven networks for trajectory prediction,” IEEE Robotics and Automation Letters, vol. 7, no. 2, pp. 2716–2723, 2022

    2022

  39. [47]

    Social-stgmlp: A social spatio- temporal graph multi-layer perceptron for pedestrian trajectory pre- diction,

    D. Meng, G. Zhao, and F. Yan, “Social-stgmlp: A social spatio- temporal graph multi-layer perceptron for pedestrian trajectory pre- diction,” Information, vol. 15, no. 6, p. 341, 2024

    2024

  40. [48]

    Evaluating pedestrian trajectory prediction methods with respect to autonomous driving,

    N. Uhlemann, F. Fent, and M. Lienkamp, “Evaluating pedestrian trajectory prediction methods with respect to autonomous driving,” IEEE Transactions on Intelligent Transportation Systems , 2024

    2024

  41. [49]

    Holistic spatio-temporal graph attention for trajectory prediction in vehicle–pedestrian interactions,

    H. Alghodhaifi and S. Lakshmanan, “Holistic spatio-temporal graph attention for trajectory prediction in vehicle–pedestrian interactions,” Sensors, vol. 23, no. 17, p. 7361, 2023

    2023

  42. [50]

    Pepscenes: A novel dataset and baseline for pedestrian action prediction in 3d,

    A. Rasouli, T. Yau, P. Lakner, S. Malekmohammadi, M. Rohani, and J. Luo, “Pepscenes: A novel dataset and baseline for pedestrian action prediction in 3d,” arXiv preprint arXiv:2012.07773 , 2020

    2012 arXiv

  43. [51]

    Openmmlab pose estimation toolbox and bench- mark

    M. Contributors, “Openmmlab pose estimation toolbox and bench- mark.” https://github.com/open-mmlab/mmpose, 2020

    2020

  44. [52]

    Analysis of pedestrian illegal crossing at unmarked segments: environmental factors, pedestrian characteristics and crossing behaviours,

    W. Zhang, H. Guo, C. Wang, K. Wang, W. Huang, Q. Xu, H. Tang, B. Yang, and R. Yan, “Analysis of pedestrian illegal crossing at unmarked segments: environmental factors, pedestrian characteristics and crossing behaviours,” Transportation research part F: traffic psychology and ...

    2023

  45. [53]

    Cou- pling intent and action for pedestrian crossing behavior prediction,

    Y . Yao, E. Atkins, M. J. Roberson, R. Vasudevan, and X. Du, “Cou- pling intent and action for pedestrian crossing behavior prediction,” arXiv preprint arXiv:2105.04133 , 2021

    2021 arXiv

  46. [54]

    Adaptive pedestrian agent model- ing for scenario-based testing of autonomous vehicles through behav- ior retargeting,

    G. M. Muktadir and J. Whitehead, “Adaptive pedestrian agent model- ing for scenario-based testing of autonomous vehicles through behav- ior retargeting,” in IEEE Int. Conf. Robot. Automat.(ICRA) , 2024

    2024

  47. [55]

    Convolutional lstm network: A machine learning approach for precipitation nowcasting,

    X. Shi, Z. Chen, H. Wang, D.-Y . Yeung, W.-K. Wong, and W.-c. Woo, “Convolutional lstm network: A machine learning approach for precipitation nowcasting,” Advances in neural information processing systems, vol. 28, 2015

    2015

  48. [56]

    RAFT: Recurrent all-pairs field transforms for optical flow,

    Z. Teed and J. Deng, “RAFT: Recurrent all-pairs field transforms for optical flow,” in Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part II 16, pp. 402–419, Springer, 2020

    2020

  49. [57]

    Social gan: Socially acceptable trajectories with generative adversarial net- works,

    A. Gupta, J. Johnson, L. Fei-Fei, S. Savarese, and A. Alahi, “Social gan: Socially acceptable trajectories with generative adversarial net- works,” in Proceedings of the IEEE conference on computer vision and pattern recognition , pp. 2255–2264, 2018

    2018

Pith tools

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