Pith. sign in

REVIEW 4 major objections 5 minor 35 references

Zero-shot Hazard Identification in Autonomous Driving: A Case Study on the COOOL Benchmark

T0 review · 4 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read A no-training pipeline cuts the COOOL hazard benchmark's baseline error by 33% and ranks 2nd of 32 teams.

desk verdict A credible but fragile competition report whose own metric definition is contradicted by its reported numbers. read the letter →

arxiv 2412.19944 v1 pith:ZFYUH3ID submitted 2024-12-27 cs.CV

Lukas Picek , Vojtěch Čermák , Marek Hanzl This is my paper · ORCID
classification cs.CV
keywords zero-shothazardidentificationautonomousdrivingCOOOLbenchmarkkernelchangepointdetectionopticalflowvision-languagemodelopen-setrecognitiondriverreaction
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 is a competition report on the COOOL benchmark for out-of-label hazards in dashcam video. It argues that a fully zero-shot pipeline—no training on the benchmark—can detect the moment a driver reacts, identify hazardous objects, and caption them, and that this pipeline reduces the official baseline's relative error by 33%. The headline result is a private-set macro accuracy of 0.51772, which placed second among 32 teams. The authors make the case that the winning combination is modular: kernel-based change-point detection run on object-size and optical-flow signals for reactions, a whitelist-exclusion filter using a small ViT classifier for hazards, and category prompting of the MOLMO vision-language model for captions.

What carries the argument

The mechanism that carries the argument is a three-part decomposition of the hazard problem into independent signals that need no task-specific training. For reactions, the core object is the kernel change-point detector applied to two time series—total bounding-box area (object size dynamic) and average optical-flow magnitude—with the first breakpoint of a fixed four-change-point setting defining the reaction moment. For identification, the core object is a whitelist-exclusion rule: a ViT-B/16 fine-tuned on CIFAR-100 produces area-weighted softmax scores, and any object whose winning class is not in a hand-selected safe-class whitelist is declared hazardous, with an additional trajectory-size filter removing near-static objects. For captioning, the core object is a prompt to MOLMO-7B asking for the five most likely class labels, with word-frequency aggregation over the five largest crops yielding the final label. None of these components is trained on COOOL data, which is the paper's central claim to zero-shot operation.

What would settle it

Evaluate the reaction detector on a held-out set with ground-truth reaction frames while sweeping the fixed change-point count from one to eight; if no count reproduces the reported 0.829 private accuracy, or if a penalty-based change-point count matches the first-breakpoint reaction at similar accuracy, the fixed-n assumption is the load-bearing and fragile part.

Watch

Extended reading notes

Core claim

The authors claim that the three COOOL sub-tasks can each be solved without any training on the benchmark's own data. Driver reactions are found by applying kernel change-point detection with an RBF kernel to the total bounding-box area per frame and to the mean dense optical-flow magnitude, then taking the first of exactly four detected breakpoints as the reaction moment; the mean ensemble of the two signals reaches a private reaction accuracy of 0.829. Hazardous objects are identified by classifying each tracked object with a ViT-B/16 fine-tuned on CIFAR-100, computing an area-weighted softmax score, keeping every class outside a five-class whitelist (pickup truck, bus, tank, motorcycle, cloud), and discarding objects whose trajectory is shorter than their width or height, reaching a private detection accuracy of 0.570. Hazard captions come from prompting MOLMO-7B to list the five most likely class labels for the five largest crops of each object, aggregating word frequencies across the crops into a final label, with a private classification accuracy of 0.162. The combined pipeline scores 0.51772 on the private set, which the authors report as a 33% relative error reduction over the baseline and second place among 32 teams.

Load-bearing premise

The pipeline assumes that fixing the number of change points to four and reading the first one as the driver reaction moment—a setting chosen on the public leaderboard—also marks the true reaction in the private videos.

Editorial extensions

If this is right

  • Driver reaction moments can be recovered from object-size and optical-flow signals alone, reaching a private-set accuracy of 0.829 with a mean-position ensemble.
  • Hazard identification can be reframed as whitelist-exclusion classification, with one small ViT reaching a private detection accuracy of 0.570.
  • Category-based prompting of a vision-language model (MOLMO) outperforms sentence captioning for low-resolution hazards, but only reaches 0.162 private classification accuracy.
  • The full pipeline reduces the COOOL baseline's relative error by 33% and secures second place on the private leaderboard.
  • On the public leaderboard, reaction accuracy drops from 0.900 (object-size dynamic) to 0.829 (private ensemble), showing that leaderboard-tuned parameters may not transfer.

Reading between the lines

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

  • A natural extension is to replace the fixed n=4 change points with a penalty-based detector (PELT or similar) and re-measure reaction accuracy; the paper's own real-time variant hints this may preserve performance without leaderboard tuning.
  • The whitelist is hand-picked from CIFAR-100 classes; a testable extension is to expand it with bicycle, pedestrian, or motorcycle-like classes and measure the trade-off between hazard recall and false positives.
  • The word-frequency aggregation could be ablated against a single MOLMO call on the largest crop; the paper reports no such comparison, so it is unclear whether the aggregation step earns its complexity.
  • Since the private-set numbers for reaction (0.829) and detection (0.570) are only slightly below the public ones, but classification drops from 0.239 to 0.162, the captioning step is the least stable component and the most promising target for domain-adaptation techniques.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. This manuscript reports the authors' submission to the COOOL benchmark for zero-shot hazard identification in autonomous driving. The pipeline combines three components: (i) driver reaction detection via Kernel Change Point Detection applied to optical flow and object-size dynamics, (ii) hazard identification based on track proximity plus a ViT-based whitelist/object classification filter, and (iii) hazard captioning using the MOLMO vision-language model with category and sentence prompts. The authors report public/private macro accuracy of 0.63993/0.51772, a 33% relative error reduction over the baseline, and a 2nd-place finish among 32 teams. The paper is presented as a competition case study rather than a general method proposal.

Significance. If the quantitative claims are correct, the paper is a useful case study showing that low-cost, zero-shot components—classical change-point detection, optical flow, and off-the-shelf vision-language models—can be competitive on a challenging out-of-distribution driving benchmark. The manuscript clearly describes each component and is honest about limitations such as low-resolution inputs, domain shift, and the absence of ground-truth validation during the competition. The principal value is the empirical comparison and the final leaderboard position, not a new algorithmic idea; hence the correctness of the evaluation is load-bearing for the paper's central claim.

major comments (4)
  1. [Section 3.2, Eq. (2) and Table 3] Under the stated metric Adetection = (1/N)Σ |Hi ∩ Ĥi|/|Hi|, predicting every annotated track as hazardous gives |Hi ∩ Ĥi| = |Hi| and hence a per-frame score of 1.0. The 'All Tracks' row in Table 3 reports 0.618/0.355, which contradicts Eq. (2). This discrepancy indicates that the detection scores—one third of the final macro score—are not computed with the metric stated in the paper. Please state the actual official metric used by the competition (e.g., a precision-sensitive or matching-based variant) and recompute or relabel the reported scores accordingly.
  2. [Sections 4.2.1 and 5.1] The number of change points n=4 and the rule that the first breakpoint marks the driver reaction are chosen based on preliminary experiments on the public leaderboard, which the paper itself describes as a validation set. The reported rank margin is very small (0.51772 vs 0.51596 for 3rd place), and a change of roughly 0.005 in Areaction could alter the final ranking. Please provide a sensitivity analysis varying n (e.g., 1–6) and the ensemble rule on the public set, and report the per-video variance of the reaction scores where possible.
  3. [Section 5.2, paragraph 2 vs Table 3] The text reads 'including all tracks achieved the highest metric score of 0.799 / 0.535', but Table 3 lists 'All Tracks' as 0.618/0.355 and assigns 0.799/0.535 to 'All + Object classification filter'. The next sentence's '+0.119' improvement is consistent with 0.355→0.474 for 'All Tracks without cars', so the 0.799/0.535 figure belongs to the filtered variant. Please correct the text to match the table.
  4. [Table 5] The header 'Apublic_reaction / Aprivate_reaction' is inconsistent with the values shown, which match the macro scores (e.g., 0.51772 for 'our' equals the reported Amacro). The reaction scores for the selected run appear in Table 1. Please relabel the columns (e.g., Apublic_macro / Aprivate_macro) and clarify which run each column refers to.
minor comments (5)
  1. [Sections 4.4 and Figure 5] The model name is spelled inconsistently as 'MOLMO', 'Malmo-7B', and 'MOMLO'; please unify the spelling throughout.
  2. [Section 4.2.3] The 'Mean Position' ensemble is described as averaging the positions of the first True; specify how the resulting mean is mapped back to a Boolean per-frame signal (e.g., rounding to the nearest frame), since the reaction metric requires per-frame labels.
  3. [Equation (3)] Please define the value of Aclassif for frames with |Ci|=0; the formula as written is undefined when the ground-truth class set is empty.
  4. [Section 3.1] The statement 'there are between 1 and 18 potential hazard objects' is ambiguous; specify whether this refers to the number of annotated objects per video or the number of true hazards per video.
  5. [Section 4.1] The CLIP-Interrogator tool and the 'Flavors' set are mentioned but the description is terse; consider a brief explanation or a direct reference to the tool's documentation.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the paper reports an empirical competition submission whose hyperparameters were tuned on the public leaderboard and evaluated on the private leaderboard, with no claim that the outputs follow from the inputs by construction.

full rationale

This is a competition report rather than a theoretical derivation, and I find no step in which a claimed prediction is equivalent to its input by construction. The pipeline components are standard, externally implemented tools (KernelCPD from ruptures, Farneback optical flow from OpenCV, a ViT-Base fine-tuned on CIFAR-100, and the MOLMO vision-language model); none of these is defined in terms of the COOOL target metric. The constants that are tuned, e.g., the n=4 change-point count in Section 4.2.1, the number of nearest tracks in Section 5.2, and the trajectory size filter in Section 5.2, were selected using the public leaderboard, which the paper explicitly acknowledges as a validation set in Section 3.1: 'we use the public leaderboard as a form of validation set.' That is a methodological limitation, not circularity, because the reported final score (0.51772) is measured on the held-out private leaderboard, which is external to the tuning signal. There is no self-citation chain carrying a load-bearing uniqueness claim, no ansatz smuggled in through a citation, and no renaming of a known result. One non-circular concern worth noting for the reader: Table 3's 'All Tracks' row (0.618 public / 0.355 private) is inconsistent with the recall-only definition in Eq. (2), since predicting every annotated track as hazardous would yield |Hi ∩ Ĥi|/|Hi| = 1.0 for every frame; this suggests the official metric may differ from the stated one, but this is a metric-description and reproducibility issue, not a circular-derivation issue. The central empirical claim therefore stands as an independent, externally evaluated result, with the usual caveat that leaderboard-tuned constants may not generalize.

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

The central results rest on three kinds of support: benchmark definitions (provided tracklets, no false-positive penalty), modeling assumptions (size/flow signals encode reactions), and leaderboard-tuned hyperparameters. No new theoretical entities are introduced.

free parameters (3)
  • n_changes_kernelcpd = 4
    Number of change points for KernelCPD on object-size and optical-flow signals, set to 4 based on preliminary experiments on the public leaderboard; the first breakpoint triggers the reaction prediction.
  • whitelist_classes = pickuptruck, bus, tank, motorcycle, cloud
    CIFAR-100 classes treated as non-hazards in the object-classification filter; chosen by observing frequent predictions, effectively tuned on the leaderboard.
  • trajectory_size_filter = remove objects with trajectory shorter than width or height
    Threshold added after observing private-leaderboard improvement (+0.035); not justified independently.
assumptions (3)
  • ad hoc to paper The first of n=4 change points in the combined object-size/optical-flow signal marks the driver reaction time.
    Used in Sections 4.2 and 5.1; if the true reaction is not aligned with the first breakpoint, the reaction accuracy collapses.
  • domain assumption All hazardous objects are present in the provided bounding-box tracklets, so hazard identification can be performed as track filtering.
    Assumed throughout Section 4.3 and visible in the evaluation, which uses the provided annotations; no detector is trained or evaluated.
  • domain assumption The evaluation metric Adetection does not penalize false positives, so marking every track as hazardous is a strong baseline.
    Evident from Eq. (2) in Section 3.2; the paper exploits this when using All Tracks as a high-scoring baseline.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Zero-shot Hazard Identification in Autonomous Driving: A Case Study on the COOOL Benchmark." pith.science (2026). https://pith.science/paper/ZFYUH3ID

@misc{pith2026241219944,
  author       = {Pith},
  title        = {Pith review of: Zero-shot Hazard Identification in Autonomous Driving: A Case Study on the COOOL Benchmark},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZFYUH3ID}},
  note         = {Machine review of arXiv:2412.19944}
}
read the original abstract

This paper presents our submission to the COOOL competition, a novel benchmark for detecting and classifying out-of-label hazards in autonomous driving. Our approach integrates diverse methods across three core tasks: (i) driver reaction detection, (ii) hazard object identification, and (iii) hazard captioning. We propose kernel-based change point detection on bounding boxes and optical flow dynamics for driver reaction detection to analyze motion patterns. For hazard identification, we combined a naive proximity-based strategy with object classification using a pre-trained ViT model. At last, for hazard captioning, we used the MOLMO vision-language model with tailored prompts to generate precise and context-aware descriptions of rare and low-resolution hazards. The proposed pipeline outperformed the baseline methods by a large margin, reducing the relative error by 33%, and scored 2nd on the final leaderboard consisting of 32 teams.

Figures

Figures reproduced from arXiv: 2412.19944 by the authors.

Figure 1
Figure 1. COOOL benchmark focus on zero-shot identification of [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Dashcam views. The COOOL dataset videos originate from different parts of the world and are of various quality. 3.1. Dataset The COOOL dataset is specifically curated to address the detection and classification of out-of-label hazards in autonomous driving. It comprises over 200 dashcam videos of relatively low quality annotated to capture diverse real-world driving scenarios (see [PITH_FULL_IMAGE:figures/full_fig_… view at source ↗
Figure 3
Figure 3. Cropped out hazard object. The COOOL dataset in￾cludes a wide variety of animals (top) and other (bottom) objects. Annotations: A bounding box and object ID annotations are provided for every frame in each available video. The dataset includes bounding box annotations covering over 100,000 vehicles, 40,000 animals, and various envi￾ronmental, miscellaneous, and unknown hazards. Each object is assigned a tracklet, en… view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Driver reaction recognition. The optical flow graph demonstrates pixel-level motion intensity over time, with spikes during significant vehicle or background movements. The Object size dynamic graph visualizes the change in bounding box sizes of hazards. Combined (i.e.…
Figure 5
Figure 5. Figure 5: Hazard captioning with MOMLO. Two prompts (1st and 3rd row) and given results. ”Correct” answer highlighted in green. 4.2.3 Ensembling To combine the scores, we explore several ensembling schemes. The signals are represented as a time series of Boolean values, starting…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

35 extracted references · 26 canonical work pages

  1. [1]

    Gpt-4 technical report

    Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ah- mad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. Gpt-4 technical report. arXiv preprint arXiv:2303.08774 ,

    arXiv

  2. [2]

    Flamingo: a visual language model for few-shot learning

    Jean-Baptiste Alayrac, Jeff Donahue, Pauline Luc, Antoine Miech, Iain Barr, Yana Hasson, Karel Lenc, Arthur Men- sch, Katherine Millican, Malcolm Reynolds, et al. Flamingo: a visual language model for few-shot learning. Advances in neural information processing systems, 35:23716–23736,

  3. [3]

    COOOL: Challenge Of Out-Of-Label A Novel Benchmark for Autonomous Driving

    Ali K AlShami, Ananya Kalita, Ryan Rabinowitz, Khang Lam, Rishabh Bezbarua, Terrance Boult, and Jugal Kalita. Coool: Challenge of out-of-label a novel benchmark for au- tonomous driving. arXiv preprint arXiv:2412.05462, 2024. 1, 2

    work page Pith review arXiv 2024

  4. [4]

    A kernel multiple change-point algorithm via model selection

    Sylvain Arlot, Alain Celisse, and Zaid Harchaoui. A kernel multiple change-point algorithm via model selection. Jour- nal of machine learning research, 20(162):1–56, 2019. 5 8

    work page 2019

  5. [5]

    Towards open set deep networks

    Abhijit Bendale and Terrance E Boult. Towards open set deep networks. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 1563–1572,

  6. [6]

    API design for machine learning soft- ware: experiences from the scikit-learn project

    Lars Buitinck, Gilles Louppe, Mathieu Blondel, Fabian Pe- dregosa, Andreas Mueller, Olivier Grisel, Vlad Niculae, Peter Prettenhofer, Alexandre Gramfort, Jaques Grobler, Robert Layton, Jake VanderPlas, Arnaud Joly, Brian Holt, and Ga¨el Varoquaux. API design for machine learning soft- ware: experiences from the scikit-learn project. In ECML PKDD Workshop:...

    work page 2013

  7. [7]

    nuscenes: A multi- modal dataset for autonomous driving

    Holger Caesar, Varun Bankiti, Alex H Lang, Sourabh V ora, Venice Erin Liong, Qiang Xu, Anush Krishnan, Yu Pan, Gi- ancarlo Baldan, and Oscar Beijbom. nuscenes: A multi- modal dataset for autonomous driving. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11621–11631, 2020. 1, 2

    work page 2020

  8. [8]

    New efficient algorithms for multiple change- point detection with reproducing kernels

    Alain Celisse, Guillemette Marot, Morgane Pierre-Jean, and GJ Rigaill. New efficient algorithms for multiple change- point detection with reproducing kernels. Computational Statistics & Data Analysis, 128:200–220, 2018. 5

    work page 2018

Show all 35 references
  1. [9]

    Recent advance- ments in end-to-end autonomous driving using deep learn- ing: A survey

    Pranav Singh Chib and Pravendra Singh. Recent advance- ments in end-to-end autonomous driving using deep learn- ing: A survey. IEEE Transactions on Intelligent Vehicles ,

  2. [10]

    Operational open-set recognition and post- max refinement

    Steve Cruz, Ryan Rabinowitz, Manuel G ¨unther, and Ter- rance E Boult. Operational open-set recognition and post- max refinement. In European Conference on Computer Vi- sion, pages 475–492. Springer, 2025. 2

    2025

  3. [11]

    Molmo and pixmo: Open weights and open data for state-of-the-art multimodal models

    Matt Deitke, Christopher Clark, Sangho Lee, Rohun Tri- pathi, Yue Yang, Jae Sung Park, Mohammadreza Salehi, Niklas Muennighoff, Kyle Lo, Luca Soldaini, et al. Molmo and pixmo: Open weights and open data for state-of-the-art multimodal models. arXiv preprint arXiv:2409.17146, 2...

    2024 arXiv

  4. [12]

    Imagenet: A large-scale hierarchical image database

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition, pages 248–255. Ieee, 2009. 6

    2009

  5. [13]

    Di Lillo, T

    L. Di Lillo, T. Gode, X. Zhou, J. M. Scanlon, R. Chen, and T. Victor. Do autonomous vehicles outperform latest-generation human-driven vehicles? a comparison to waymo’s auto liability insurance claims at 25 million miles,

  6. [14]

    Trust your generator (trygen): Enhancing out-of-model scope detection

    V ´aclav Diviˇs, Bastian Spatz, and Marek Hr ´uz. Trust your generator (trygen): Enhancing out-of-model scope detection. AI, 5(4):2127–2146, 2024. 2

    2024

  7. [15]

    Two-frame motion estimation based on polynomial expansion

    Gunnar Farneb ¨ack. Two-frame motion estimation based on polynomial expansion. In Image Analysis: 13th Scandina- vian Conference, SCIA 2003 Halmstad, Sweden, June 29– July 2, 2003 Proceedings 13, pages 363–370. Springer, 2003. 5

    2003

  8. [16]

    A re- view on deep learning techniques applied to semantic seg- mentation

    Alberto Garcia-Garcia, Sergio Orts-Escolano, Sergiu Oprea, Victor Villena-Martinez, and Jose Garcia-Rodriguez. A re- view on deep learning techniques applied to semantic seg- mentation. arXiv preprint arXiv:1704.06857, 2017. 1, 2

    2017 arXiv

  9. [17]

    A baseline for detect- ing misclassified and out-of-distribution examples in neural networks

    Dan Hendrycks and Kevin Gimpel. A baseline for detect- ing misclassified and out-of-distribution examples in neural networks. In International Conference on Learning Repre- sentations, 2017. 2

    2017

  10. [18]

    Evaluation of out-of-distribution detection perfor- mance on autonomous driving datasets

    Jens Henriksson, Christian Berger, Stig Ursing, and Markus Borg. Evaluation of out-of-distribution detection perfor- mance on autonomous driving datasets. In 2023 IEEE International Conference On Artificial Intelligence Testing (AITest), pages 74–81. IEEE, 2023. 2

    2023

  11. [19]

    Planning-oriented autonomous driving

    Yihan Hu, Jiazhi Yang, Li Chen, Keyu Li, Chonghao Sima, Xizhou Zhu, Siqi Chai, Senyao Du, Tianwei Lin, Wenhai Wang, et al. Planning-oriented autonomous driving. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 17853–17862, 2023. 1

    2023

  12. [20]

    Open source computer vision library

    Itseez. Open source computer vision library. https:// github.com/itseez/opencv, 2015. 5

    2015

  13. [21]

    Op- timal detection of changepoints with a linear computa- tional cost

    Rebecca Killick, Paul Fearnhead, and Idris A Eckley. Op- timal detection of changepoints with a linear computa- tional cost. Journal of the American Statistical Association, 107(500):1590–1598, 2012. 5

    2012

  14. [22]

    Cifar- 100 (canadian institute for advanced research)

    Alex Krizhevsky, Vinod Nair, and Geoffrey Hinton. Cifar- 100 (canadian institute for advanced research). 6, 7

  15. [23]

    Large car-following data based on lyft level-5 open dataset: Following autonomous vehicles vs

    Guopeng Li, Yiru Jiao, Victor L Knoop, Simeon C Calvert, and JWC Van Lint. Large car-following data based on lyft level-5 open dataset: Following autonomous vehicles vs. human-driven vehicles. In 2023 IEEE 26th International Conference on Intelligent Transportation Systems (IT...

    2023

  16. [24]

    Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation

    Junnan Li, Dongxu Li, Caiming Xiong, and Steven Hoi. Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation. In Interna- tional conference on machine learning, pages 12888–12900. PMLR, 2022. 4, 6

    2022

  17. [25]

    Learning transferable visual models from natural language supervi- sion

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervi- sion. In International conference on machine learning, ...

    2021

  18. [26]

    Deep learning serves traffic safety analysis: A forward-looking review

    Abolfazl Razi, Xiwen Chen, Huayu Li, Hao Wang, Brendan Russo, Yan Chen, and Hongbin Yu. Deep learning serves traffic safety analysis: A forward-looking review. IET Intel- ligent Transport Systems, 17(1):22–71, 2023. 1

    2023

  19. [27]

    The extreme value machine

    Ethan M Rudd, Lalit P Jain, Walter J Scheirer, and Ter- rance E Boult. The extreme value machine. IEEE transactions on pattern analysis and machine intelligence , 40(3):762–768, 2017. 2

    2017

  20. [28]

    Scalability in perception for autonomous driving: Waymo open dataset

    Pei Sun, Henrik Kretzschmar, Xerxes Dotiwalla, Aurelien Chouard, Vijaysai Patnaik, Paul Tsui, James Guo, Yin Zhou, Yuning Chai, Benjamin Caine, et al. Scalability in perception for autonomous driving: Waymo open dataset. In Proceed- ings of the IEEE/CVF conference on computer ...

    2020

  21. [29]

    Selec- tive review of offline change point detection methods.Signal Processing, 167:107299, 2020

    Charles Truong, Laurent Oudre, and Nicolas Vayatis. Selec- tive review of offline change point detection methods.Signal Processing, 167:107299, 2020. 5 9

    2020

  22. [30]

    Image captioners sometimes tell more than images they see

    Honori Udo and Takafumi Koshinaka. Image captioners sometimes tell more than images they see. arXiv preprint arXiv:2305.02932, 2023. 4

    2023 arXiv

  23. [31]

    Open-set recognition: A good closed-set classifier is all you need

    Sagar Vaze, Kai Han, Andrea Vedaldi, and Andrew Zisser- man. Open-set recognition: A good closed-set classifier is all you need. In International Conference on Learning Rep- resentations, 2022. 2

    2022

  24. [32]

    Pixood: Pixel- level out-of-distribution detection

    Tom ´aˇs V oj´ıˇr, Jan ˇSochman, and Ji ˇr´ı Matas. Pixood: Pixel- level out-of-distribution detection. In European Conference on Computer Vision, pages 93–109. Springer, 2025. 2

    2025

  25. [33]

    Visual transform- ers: Token-based image representation and processing for computer vision, 2020

    Bichen Wu, Chenfeng Xu, Xiaoliang Dai, Alvin Wan, Peizhao Zhang, Zhicheng Yan, Masayoshi Tomizuka, Joseph Gonzalez, Kurt Keutzer, and Peter Vajda. Visual transform- ers: Token-based image representation and processing for computer vision, 2020. 7

    2020

  26. [34]

    Autonomous driving system: A comprehensive survey

    Jingyuan Zhao, Wenyi Zhao, Bo Deng, Zhenghong Wang, Feng Zhang, Wenxiang Zheng, Wanke Cao, Jinrui Nan, Yubo Lian, and Andrew F Burke. Autonomous driving system: A comprehensive survey. Expert Systems with Applications , 242:122836, 2024. 1 10

    2024

  27. [2024]

    Waymo Research Blog. 1

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.