Pith. sign in

REVIEW 3 major objections 6 minor 45 references

S2R-Bench: A Sim-to-Real Evaluation Benchmark for Autonomous Driving

T0 review · 3 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read This paper introduces S2R-Bench, which it claims is the first corruption-robustness benchmark built from real-world autonomous-driving sensor data, pairing real adverse-weather sequences with simulated corruptions to test whether…

desk verdict Useful real-world multi-sensor anomaly dataset, but the LiDAR-based labels under snow undermine the paper's validation claims until an annotation audit is provided. read the letter →

arxiv 2505.18631 v1 pith:3TUEB3C2 submitted 2025-05-24 cs.RO

Li Wang , Guangqi Yang , Lei Yang , Ziying Song , Xinyu Zhang , Ying Chen , Lin Liu , Junjie Gao
show 8 more authors
Zhiwei Li Qingshan Yang Jun Li Liangliang Wang Wenhao Yu Bin Xu Weida Wang Huaping Liu
This is my paper · ORCID
classification cs.RO
keywords autonomousdrivingcorruptionrobustnessbenchmarksim-to-realevaluationsensoranomaliesadverseweather4Dradar3Dobjectdetectionmultimodalperception
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 introduces S2R-Bench, a benchmark for evaluating how well autonomous-driving perception systems survive sensor anomalies. The authors claim it is the first corruption-robustness benchmark built from real-world driving data rather than purely simulated corruptions, and they collect 10,117 synchronized frames from a camera, an 80-line LiDAR, and two 4D radars under light snow, moderate snow, fog, brightness, tunnel, and misalignment conditions. Each real sequence is paired with a simulated counterpart produced by four existing corruption suites, so the same detector can be scored on real and simulated versions of the same scenario. The benchmark's purpose is to show whether simulated corruptions predict real-world failures and to provide a realistic testbed for improving perception robustness.

What carries the argument

The carrying object is the paired dataset itself: S2R-C provides clean sequences, S2R-R provides real sensor anomaly sequences, and S2R-S provides simulated anomaly sequences generated from clean data by four corruption suites (3D Corruptions AD, MultiCorrupt, Robo3D, and RoboDepth, the last image-only). The comparison protocol is to train a detector on S2R-C and measure its performance on S2R-R versus S2R-S, using LiDAR-derived 3D bounding boxes projected onto the camera and radar data as common annotations; a scenario counts as well simulated when the real and simulated scores are within 30% of each other. This protocol is what lets the benchmark turn a collection of difficult weather clips into an external-validation instrument for simulated corruption methods.

What would settle it

Independently re-annotate a random sample of snowy S2R-R frames using a LiDAR-independent source such as manual human annotation from high-resolution images or dense radar reflectivity, then compare object recall and box overlap with the released LiDAR-projected labels; if recall falls sharply as snow intensity rises or many boxes shift beyond the dataset's own tolerance, the central comparison rests on biased ground truth.

Watch

Extended reading notes

Core claim

The paper's central claim is that S2R-Bench makes it possible to validate simulated sensor corruptions against real ones: models trained on clean data are tested on real adverse-weather sequences (S2R-R) and on simulated versions (S2R-S), and the paper treats a scenario as well simulated when the real and simulated performance differ by less than 30%. Under that criterion, the reported experiments with three detection baselines show that no single simulation method is faithful everywhere: certain methods approximate real fog or light-snow behavior for particular sensors, while spatial misalignment and brightness anomalies remain poorly modeled. From this comparison the paper concludes that the collected real-world data is reliable and practically significant as an evaluation target, and that the benchmark can expose where current simulation-based robustness tests diverge from reality.

Load-bearing premise

The load-bearing premise is that the LiDAR-derived 3D bounding boxes are correct ground truth in every scenario, even though the paper itself reports that snowflakes add significant noise to LiDAR; if those labels miss or misplace objects in snowy frames, every real-versus-simulated comparison built on them is compromised.

Editorial extensions

If this is right

  • Perception models can be ranked on real snow, fog, brightness, tunnel, and misalignment data instead of only on synthetic corruptions, giving a more direct estimate of deployment-time failure.
  • The results imply that simulation-based robustness benchmarks should be treated as scenario-dependent: some corruption suites match real behavior for some sensors and weather types, while misalignment and brightness simulations are consistently unreliable.
  • The synchronized camera-LiDAR-4D-radar data allow researchers to study which sensors degrade most under each condition; the paper notes LiDAR picks up heavy snowflake noise while the 4D radar remains comparatively stable.
  • Beyond 3D object detection, the conclusion states the dataset supports anomaly detection, sensor fusion, and generalization research.

Reading between the lines

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

  • A fair reading of the reported numbers suggests that an aggregate robustness ranking across all scenarios would be misleading: simulation fidelity is sensor-specific and corruption-type-specific, so future benchmarks should report per-sensor, per-scenario breakdowns rather than one score.
  • The 30% good-simulation threshold is an authorial choice; repeating the experiments with multiple training seeds and reporting confidence intervals could change which methods are judged faithful, especially where differences are near the boundary.
  • Because the collection window covers one winter in one city, the relative ordering of simulation methods might not transfer to rain, hail, or road-surface changes; extending the benchmark to other climates and seasons is a direct way to test how general the sim-to-real conclusions are.
  • The LiDAR-based label pipeline could itself be the source of part of the real-versus-simulated gap: if snowflake noise makes LiDAR miss objects, then detector failures on S2R-R might partly reflect label bias rather than genuine robustness differences.
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

3 major / 6 minor

Summary. The paper introduces S2R-Bench, a multimodal benchmark for evaluating perception robustness of autonomous driving under real-world sensor anomalies and for comparing real data with simulated corruptions. It comprises three components: S2R-R (real sensor anomaly data collected in Beijing under light/moderate snow, fog, brightness, tunnel, misalignment, and blur conditions), S2R-C (clean data), and S2R-S (simulated anomalies generated by four existing corruption methods). Using a camera, an 80-line LiDAR, and two 4D radars, the dataset includes 151 sequences, 10,117 synchronized frames, and 3D bounding box annotations. The authors train PointPillars, SMOKE, and Focals Conv and compare their performance on real versus simulated test data, defining a within-30% variation as a 'good simulation effect.' The paper claims that this is the first corruption robustness benchmark based on real-world scenarios and that the comparisons demonstrate the reliability of the real data for practical autonomous driving applications.

Significance. If the dataset and annotations were properly validated, S2R-Bench would be a valuable community resource: it is among the first multi-modal (camera, LiDAR, 4D radar) benchmarks containing real adverse-weather anomaly data alongside corresponding simulated data, and it makes concrete comparisons across three baseline architectures. The paper ships a released dataset and evaluation code, which is a positive contribution. However, the central claim that the real data are reliable can serve as a reference for validating simulation methods rests on three load-bearing assumptions: (i) the LiDAR-derived ground truth remains accurate in snowy and foggy conditions where the paper itself notes LiDAR is heavily corrupted, (ii) single-run AP numbers without error bars or significance tests are sufficient to support the stated quantitative conclusions, and (iii) the within-30% criterion is a meaningful definition of simulation fidelity. These weaknesses currently make the quantitative results suggestive rather than conclusive, although they are addressable with additional validation.

major comments (3)
  1. [III-B / II-F] The benchmark's central claim that real-world data can validate simulations depends on the reliability of the S2R-R ground truth. Section III-B states that annotations are 'based on the LiDAR point cloud data,' while Section II-F and Fig. 5 note that in snowy scenarios 'LiDAR is particularly prone to interference from snowflakes, introducing significant noise' and that 4D radar is 'less affected by weather.' Since the snowy subset comprises 2,206 light-snow and 1,198 moderate-snow frames, LiDAR-based boxes are likely incomplete or biased in exactly the conditions where the real-versus-simulated comparison is most important. Radar points without a LiDAR-derived box may be scored as false positives, penalizing the sensor least affected by weather. The paper reports no inter-annotator agreement, no miss-rate audit, and no independent label check (e.g., using 4D radar returns). Without such an audit, every real-world AP value in Tables II-X and every within-30% conclusion is computed against an unverified reference.
  2. [IV, Tables II-X] All experimental tables report single-run AP values with no error bars, confidence intervals, or significance tests. The text repeatedly asserts that a simulation method 'closely approximates' real data (e.g., Section IV-A compares real Fog LiDAR 3D AP of 37.63 with simulated 24.42, a 35% gap, yet Section IV-B claims 'identical accuracy' for Robo3D in the Fog row of Table V); these claims are not statistically supported. Many entries are identical across rows and modalities (e.g., the Robo3D row in Table III reports 38.78 for every scenario column), suggesting that some simulations produce degenerate or constant outputs, which is not discussed. Without repeated runs or a bootstrap analysis, the paper's core quantitative evidence for the reliability of real data is inconclusive.
  3. [IV] The criterion 'data variation within 30%' is introduced in Section IV without justification or formal definition. It is not derived from task requirements, statistical considerations, or prior benchmarks; no rationale is given for why 30% rather than some other threshold signifies a 'good simulation effect.' The bold highlighting based on this threshold therefore does not constitute evidence of sim-to-real fidelity. The authors should either justify the threshold with a principled argument (e.g., calibration to downstream safety requirements) or replace it with a continuous measure of agreement with uncertainty quantification.
minor comments (6)
  1. [Abstract / I] The abstract and introduction contain grammatical errors and duplicated phrases, for example 'entierly simulated,' 'the use of benchmarks are entierly simulated,' and the repeated sentence 'Therefore, investigating the robustness of autonomous driving perception...' in Section I; these should be corrected.
  2. [II-D] The dataset organization description is confusing: it states that S2R-S consists of simulated data for each S2R-R scenario but then says the simulation includes four sequences representing daytime and nighttime conditions in S2R-C; the relationship between S2R-S and S2R-R/S2R-C should be clarified.
  3. [II-E, Tables II-III] Table II's caption lists three subsets of S2R-S while Table III lists four; the text in Section II-E mentions four simulation methods but does not explain that RoboDepth is omitted from Table II because it only simulates image data. The captions should state this explicitly.
  4. [II-F, Figs. 5-6] The description of the odd/even rows in Section II-F is garbled and contains a duplicated sentence; additionally, the spelling of the radar sensor name is inconsistent ('Oculli-Eagle' in Table I versus 'Oculii-Eagle' in the text and figure captions).
  5. [V] The conclusion claims that S2R-Bench supports anomaly detection, sensor fusion, and generalization research, but no experiments for those downstream tasks are presented; this claim should be softened to a statement of potential usefulness.
  6. [I] The paper's novelty claim of being 'the first corruption robustness benchmark based on real-world scenarios' is not positioned against existing real-world adverse-weather datasets such as K-Radar or SeeingThroughFog; the claim should be tempered or differentiated.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: S2R-Bench's sim-to-real comparison is an empirical evaluation against external baselines and published simulation methods, not a derivation that reduces to its own inputs.

full rationale

S2R-Bench is a data-collection and evaluation benchmark, not a derived prediction. Its central comparison—training detectors on S2R-C and measuring AP on real anomalies (S2R-R) versus simulated anomalies (S2R-S)—uses standard external baselines (PointPillars, SMOKE, Focals Conv) and externally published corruption methods (KITTI-C, MultiCorrupt, Robo3D, RoboDepth). No parameter is fitted to a subset of the anomaly data and then presented as a prediction; the 'within 30%' rule is an explicitly stated interpretation threshold (Section IV), not a fitted constant. The dataset links in refs [27]–[31] are self-references but only point to the released artifact and are not load-bearing for the experimental conclusions. The LiDAR-based annotation protocol (Section III-B) creates a real validity risk for radar evaluation in snow, given the paper's own statement that 'LiDAR is particularly prone to interference from snowflakes' (Section II-F); however, this is an annotation-quality and correctness concern, not a circular reduction, because the benchmark's AP comparisons are not defined in terms of the conclusions they support. No self-citation chain or uniqueness theorem is invoked to force the paper's design choices. Therefore no circular step can be exhibited, and the score is 0.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The benchmark introduces dataset partitions (S2R-C, S2R-R, S2R-S) but no new physical entities. The main hidden burden is the reliance on LiDAR-based annotations under conditions that corrupt LiDAR, plus the arbitrary 30% match threshold used to interpret results.

free parameters (1)
  • Simulation-match threshold = 30%
    Section IV defines 'cases where the data variation is within 30%' as indicating good simulation performance. This threshold is chosen by hand, has no statistical justification, and directly determines which results are highlighted as supporting the benchmark's reliability.
assumptions (4)
  • domain assumption LiDAR point clouds provide reliable ground truth in adverse weather.
    Section III-B states that all 3D annotations are based on LiDAR point clouds and projected to other sensors. Section II-B and Figure 5 acknowledge that LiDAR is significantly corrupted by snowflakes and fog, so this assumption may be false.
  • domain assumption Existing corruption simulation methods are a representative proxy for real sensor anomalies.
    Section III-A generates all S2R-S data using KITTI-C, MultiCorrupt, Robo3D, and RoboDepth. The conclusion that simulated data can be used to evaluate real-world robustness depends on this assumption.
  • domain assumption Data collected in Beijing during one winter generalizes to other regions, seasons, and sensor configurations.
    Section II-B describes data collection from December 2023 to January 2024 in Beijing using one vehicle platform. The paper frames S2R-Bench as a general-purpose benchmark without discussing generalization limits.
  • domain assumption AP at medium difficulty for the car category is a sufficient measure of perception robustness.
    Section IV reports results only for the 'car' category at medium difficulty. Pedestrian and cyclist robustness, which are safety-critical, are not evaluated, so the benchmark's claims about safety-relevant robustness are incomplete.

how reviews work

0 comments
Cite this review

Pith. "Pith review of S2R-Bench: A Sim-to-Real Evaluation Benchmark for Autonomous Driving." pith.science (2026). https://pith.science/paper/3TUEB3C2

@misc{pith2026250518631,
  author       = {Pith},
  title        = {Pith review of: S2R-Bench: A Sim-to-Real Evaluation Benchmark for Autonomous Driving},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3TUEB3C2}},
  note         = {Machine review of arXiv:2505.18631}
}
read the original abstract

Safety is a long-standing and the final pursuit in the development of autonomous driving systems, with a significant portion of safety challenge arising from perception. How to effectively evaluate the safety as well as the reliability of perception algorithms is becoming an emerging issue. Despite its critical importance, existing perception methods exhibit a limitation in their robustness, primarily due to the use of benchmarks are entierly simulated, which fail to align predicted results with actual outcomes, particularly under extreme weather conditions and sensor anomalies that are prevalent in real-world scenarios. To fill this gap, in this study, we propose a Sim-to-Real Evaluation Benchmark for Autonomous Driving (S2R-Bench). We collect diverse sensor anomaly data under various road conditions to evaluate the robustness of autonomous driving perception methods in a comprehensive and realistic manner. This is the first corruption robustness benchmark based on real-world scenarios, encompassing various road conditions, weather conditions, lighting intensities, and time periods. By comparing real-world data with simulated data, we demonstrate the reliability and practical significance of the collected data for real-world applications. We hope that this dataset will advance future research and contribute to the development of more robust perception models for autonomous driving. This dataset is released on https://github.com/adept-thu/S2R-Bench.

Figures

Figures reproduced from arXiv: 2505.18631 by the authors.

Figure 1
Figure 1. The impact of diverse adverse weather conditions and driving scenarios on camera lens clarity and captured image quality in autonomous driving [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The configuration of our experiment platform and visualisation scenarios. (a) Shows the self-driving car system as well as the real anomaly data [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Projection visualization of sensor calibration. (a), (b), (c), and (d) represent the projection of the calibrated data (3D bounding box, LiDAR point [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: The statistic of different annotated objects at [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Representing 3D annotations in multiple scenarios and sensor modalities. The four columns respectively display the projection of 3D annotation boxes [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Representing 3D annotations in multiple scenarios and sensor modalities. Each row represents one scenario. (i) tunnel [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

45 extracted references · 40 canonical work pages

  1. [1]

    Online legal driving behavior monitoring for self-driving vehicles,

    W. Yu, C. Zhao, H. Wang, J. Liu, X. Ma, Y . Yang, J. Li, W. Wang, X. Hu, and D. Zhao, “Online legal driving behavior monitoring for self-driving vehicles,” Nature communications, vol. 15, no. 1, p. 408, 2024

    work page 2024

  2. [2]

    Toward robust 3d perception for autonomous vehicles: A review of adversarial attacks and countermeasures,

    K. Y . Mahima, A. G. Perera, S. Anavatti, and M. Garratt, “Toward robust 3d perception for autonomous vehicles: A review of adversarial attacks and countermeasures,” IEEE Transactions on Intelligent Transportation Systems, 2024

    work page 2024

  3. [3]

    Benchmarking object detection robustness against real-world corruptions,

    J. Liu, Z. Wang, L. Ma, C. Fang, T. Bai, X. Zhang, J. Liu, and Z. Chen, “Benchmarking object detection robustness against real-world corruptions,” International Journal of Computer Vision, pp. 1–19, 2024

    work page 2024

  4. [4]

    Fear-neuro-inspired reinforcement learning for safe autonomous driving,

    X. He, J. Wu, Z. Huang, Z. Hu, J. Wang, A. Sangiovanni-Vincentelli, and C. Lv, “Fear-neuro-inspired reinforcement learning for safe autonomous driving,” IEEE transactions on pattern analysis and machine intelli- gence, 2023

    work page 2023

  5. [5]

    Rcbevdet: Radar-camera fusion in bird’s eye view for 3d object detection,

    Z. Lin, Z. Liu, Z. Xia, X. Wang, Y . Wang, S. Qi, Y . Dong, N. Dong, L. Zhang, and C. Zhu, “Rcbevdet: Radar-camera fusion in bird’s eye view for 3d object detection,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2024, pp. 14 928–14 937

    work page 2024

  6. [6]

    Spatio-spectral 4d coherent ranging using a flutter-wavelength- swept laser,

    D. Jeong, H. Jang, M. U. Jung, T. Jeong, H. Kim, S. Yang, J. Lee, and C.- S. Kim, “Spatio-spectral 4d coherent ranging using a flutter-wavelength- swept laser,” Nature Communications, vol. 15, no. 1, p. 1110, 2024

    work page 2024

  7. [7]

    Mufasa: Multi-view fusion and adaptation network with spatial aware- ness for radar object detection,

    X. Peng, M. Tang, H. Sun, K. Bierzynski, L. Servadei, and R. Wille, “Mufasa: Multi-view fusion and adaptation network with spatial aware- ness for radar object detection,” in International Conference on Artificial Neural Networks. Springer, 2024, pp. 168–184

    work page 2024

  8. [8]

    Photonic radar for contactless vital sign detection,

    Z. Zhang, Y . Liu, T. Stephens, and B. J. Eggleton, “Photonic radar for contactless vital sign detection,” Nature Photonics, vol. 17, no. 9, pp. 791–797, 2023

    work page 2023

Show all 45 references
  1. [9]

    Vision meets robotics: The kitti dataset,

    A. Geiger, P. Lenz, C. Stiller, and R. Urtasun, “Vision meets robotics: The kitti dataset,” The International Journal of Robotics Research , vol. 32, no. 11, pp. 1231–1237, 2013

    2013

  2. [10]

    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, 2020, pp. 11 621–1...

    2020

  3. [11]

    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, 2...

    2020

  4. [12]

    K-radar: 4d radar object detection for autonomous driving in various weather conditions,

    D.-H. Paek, S.-H. Kong, and K. T. Wijaya, “K-radar: 4d radar object detection for autonomous driving in various weather conditions,” Ad- vances in Neural Information Processing Systems , vol. 35, pp. 3819– 3829, 2022

    2022

  5. [13]

    Benchmarking robustness of 3d object detection to common corruptions,

    Y . Dong, C. Kang, J. Zhang, Z. Zhu, Y . Wang, X. Yang, H. Su, X. Wei, and J. Zhu, “Benchmarking robustness of 3d object detection to common corruptions,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2023, pp. 1022–1032

    2023

  6. [14]

    How do weather events impact roads: https://ops.fhwa.dot.gov/weather/ q1 roadimpact.htm,

    “How do weather events impact roads: https://ops.fhwa.dot.gov/weather/ q1 roadimpact.htm,” 2016

    2016

  7. [15]

    A noise analysis of 4d radar: robust sensing for automotive?

    P. H. Chan, S. S. Roudposhti, X. Ye, and V . Donzella, “A noise analysis of 4d radar: robust sensing for automotive?” Authorea Preprints, vol. 14, 2023

    2023

  8. [16]

    Benchmarking the robustness of semantic segmentation models,

    C. Kamann and C. Rother, “Benchmarking the robustness of semantic segmentation models,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , 2020, pp. 8828–8838

    2020

  9. [17]

    Towards viewpoint robustness in bird’s eye view segmentation,

    T. Klinghoffer, J. Philion, W. Chen, O. Litany, Z. Gojcic, J. Joo, R. Raskar, S. Fidler, and J. M. Alvarez, “Towards viewpoint robustness in bird’s eye view segmentation,” in Proceedings of the IEEE/CVF International Conference on Computer Vision , 2023, pp. 8515–8524

    2023

  10. [18]

    Understanding the robustness in vision transformers,

    D. Zhou, Z. Yu, E. Xie, C. Xiao, A. Anandkumar, J. Feng, and J. M. Alvarez, “Understanding the robustness in vision transformers,” in International Conference on Machine Learning . PMLR, 2022, pp. 27 378–27 394

    2022

  11. [19]

    Understanding the robustness of 3d object detection with bird’s-eye-view representations in autonomous driving,

    Z. Zhu, Y . Zhang, H. Chen, Y . Dong, S. Zhao, W. Ding, J. Zhong, and S. Zheng, “Understanding the robustness of 3d object detection with bird’s-eye-view representations in autonomous driving,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni- t...

    2023

  12. [20]

    Robo3d: Towards robust and reliable 3d perception against corruptions,

    L. Kong, Y . Liu, X. Li, R. Chen, W. Zhang, J. Ren, L. Pan, K. Chen, and Z. Liu, “Robo3d: Towards robust and reliable 3d perception against corruptions,” in Proceedings of the IEEE/CVF International Conference on Computer Vision , 2023, pp. 19 994–20 006

    2023

  13. [21]

    Ro- bodepth: Robust out-of-distribution depth estimation under corruptions,

    L. Kong, S. Xie, H. Hu, L. X. Ng, B. Cottereau, and W. T. Ooi, “Ro- bodepth: Robust out-of-distribution depth estimation under corruptions,” Advances in Neural Information Processing Systems , vol. 36, 2024

    2024

  14. [22]

    A flexible new technique for camera calibration,

    Z. Zhengyou, “A flexible new technique for camera calibration,” Mi- crosoft Research Technical Report, 1998

    1998

  15. [23]

    Automatic cam- era and range sensor calibration using a single shot,

    A. Geiger, F. Moosmann, ¨O. Car, and B. Schuster, “Automatic cam- era and range sensor calibration using a single shot,” in 2012 IEEE international conference on robotics and automation . IEEE, 2012, pp. 3936–3943

    2012

  16. [24]

    Radar and vision sensors calibration for outdoor 3d reconstruction,

    G. El Natour, O. A. Aider, R. Rouveure, F. Berry, and P. Faure, “Radar and vision sensors calibration for outdoor 3d reconstruction,” in 2015 IEEE International Conference on Robotics and Automation (ICRA) . IEEE, 2015, pp. 2084–2089

    2015

  17. [25]

    Lidar- camera calibration using 3d-3d point correspondences,

    A. Dhall, K. Chelani, V . Radhakrishnan, and K. M. Krishna, “Lidar- camera calibration using 3d-3d point correspondences,” arXiv preprint arXiv:1705.09785, 2017

    2017 arXiv

  18. [26]

    Extrinsic 6dof calibration of a radar–lidar–camera system enhanced by radar cross section estimates evaluation,

    J. Per ˇsi´c, I. Markovi ´c, and I. Petrovi ´c, “Extrinsic 6dof calibration of a radar–lidar–camera system enhanced by radar cross section estimates evaluation,” Robotics and Autonomous Systems , vol. 114, pp. 217–230, 2019

    2019

  19. [27]

    S2r-bench: Dataset clean part,

    W. Li, “S2r-bench: Dataset clean part,” Figshare, 2024, https://figshare.com/s/dd43a2f00fa5f332496c

    2024

  20. [28]

    S2r-bench: Dataset real part01,

    ——, “S2r-bench: Dataset real part01,” Figshare, 2024, https://figshare.com/s/02cc7ee29a64d65b8fcc

    2024

  21. [29]

    S2r-bench: Dataset real part02,

    ——, “S2r-bench: Dataset real part02,” Figshare, 2024, https://figshare.com/s/5c6d2eaa11e31e79f0b4

    2024

  22. [30]

    S2r-bench: Dataset simulation part01,

    ——, “S2r-bench: Dataset simulation part01,” Figshare, 2024, https://figshare.com/s/034cac69a46e166898a7

    2024

  23. [31]

    S2r-bench: Dataset simulation part02,

    ——, “S2r-bench: Dataset simulation part02,” Figshare, 2024, https://figshare.com/s/cd38fb2d9d1e23a0dd44

    2024

  24. [32]

    Multicorrupt: A multi- modal robustness dataset and benchmark of lidar-camera fusion for 3d object detection,

    T. Beemelmanns, Q. Zhang, and L. Eckstein, “Multicorrupt: A multi- modal robustness dataset and benchmark of lidar-camera fusion for 3d object detection,” arXiv preprint arXiv:2402.11677 , 2024

    2024 arXiv

  25. [33]

    Lidar snowfall simulation for robust 3d object detection,

    M. Hahner, C. Sakaridis, M. Bijelic, F. Heide, F. Yu, D. Dai, and L. Van Gool, “Lidar snowfall simulation for robust 3d object detection,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 16 364–16 374

    2022

  26. [34]

    Understanding lidar performance for autonomous vehicles under snowfall conditions,

    C. Sun, P. Sun, J. Wang, Y . Guo, and X. Zhao, “Understanding lidar performance for autonomous vehicles under snowfall conditions,” IEEE Transactions on Intelligent Transportation Systems , pp. 1–11, 2024

    2024

  27. [35]

    Research on the visual image-based complexity perception method of autonomous navigation scenes for unmanned surface vehicles,

    B. Shi, J. Guo, C. Wang, Y . Su, Y . Di, and M. S. AbouOmar, “Research on the visual image-based complexity perception method of autonomous navigation scenes for unmanned surface vehicles,” Scientific Reports , vol. 12, no. 1, p. 10370, 2022

    2022

  28. [36]

    Domain adaptive object detection for autonomous driving under foggy weather,

    J. Li, R. Xu, J. Ma, Q. Zou, J. Ma, and H. Yu, “Domain adaptive object detection for autonomous driving under foggy weather,” in Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2023, pp. 612–622

    2023

  29. [37]

    Perception and sensing for autonomous vehicles under adverse weather conditions: A survey,

    Y . Zhang, A. Carballo, H. Yang, and K. Takeda, “Perception and sensing for autonomous vehicles under adverse weather conditions: A survey,” ISPRS Journal of Photogrammetry and Remote Sensing , vol. 196, pp. 146–177, 2023

    2023

  30. [38]

    Automatic extrinsic calibration for lidar-stereo vehicle sensor setups,

    C. Guindel, J. Beltr ´an, D. Mart ´ın, and F. Garc ´ıa, “Automatic extrinsic calibration for lidar-stereo vehicle sensor setups,” in 2017 IEEE 20th international conference on intelligent transportation systems (ITSC) . IEEE, 2017, pp. 1–6

    2017

  31. [39]

    Automatic clarity threshold determination of blurred images for computer vision tasks’ security,

    N. Li, K. Wu, C. Gu, and T. Guan, “Automatic clarity threshold determination of blurred images for computer vision tasks’ security,” in 2021 6th International Conference on Computational Intelligence and Applications (ICCIA). IEEE, 2021, pp. 190–194

    2021

  32. [40]

    On failures of rgb cameras and their effects in autonomous driving applications,

    F. Secci and A. Ceccarelli, “On failures of rgb cameras and their effects in autonomous driving applications,” in 2020 IEEE 31st International Symposium on Software Reliability Engineering (ISSRE) . IEEE, 2020, pp. 13–24

    2020

  33. [41]

    An automatic lane marking detection method with low-density roadside lidar data,

    C. Lin, Y . Guo, W. Li, H. Liu, and D. Wu, “An automatic lane marking detection method with low-density roadside lidar data,” IEEE Sensors Journal, vol. 21, no. 8, pp. 10 029–10 038, 2021

    2021

  34. [42]

    An application-driven concep- tualization of corner cases for perception in highly automated driving,

    F. Heidecker, J. Breitenstein, K. R ¨osch, J. L ¨ohdefink, M. Bieshaar, C. Stiller, T. Fingscheidt, and B. Sick, “An application-driven concep- tualization of corner cases for perception in highly automated driving,” in 2021 IEEE Intelligent Vehicles Symposium (IV) . IEEE, 202...

    2021

  35. [43]

    Pointpillars: Fast encoders for object detection from point clouds,

    A. H. Lang, S. V ora, H. Caesar, L. Zhou, J. Yang, and O. Beijbom, “Pointpillars: Fast encoders for object detection from point clouds,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019, pp. 12 697–12 705

    2019

  36. [44]

    Smoke: Single-stage monocular 3d object detection via keypoint estimation,

    Z. Liu, Z. Wu, and R. T ´oth, “Smoke: Single-stage monocular 3d object detection via keypoint estimation,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition workshops, 2020, pp. 996–997

    2020

  37. [45]

    Focal sparse convolutional networks for 3d object detection,

    Y . Chen, Y . Li, X. Zhang, J. Sun, and J. Jia, “Focal sparse convolutional networks for 3d object detection,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2022, pp. 5428–5437

    2022

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.