Pith. sign in

REVIEW 3 major objections 5 minor 66 references

What Really Matters for Learning-based LiDAR-Camera Calibration

T0 review · 3 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read Regression-based LiDAR-camera calibration networks do not match image and point cloud features; they memorize a depth-map-to-pose mapping that fails on any other sensor setup.

desk verdict A useful diagnostic paper arguing regression-based calibration nets are retrieval nets; the claim is plausible but needs an intrinsic-controlled test and code release. read the letter →

arxiv 2501.16969 v1 pith:Q4JEFBRH submitted 2025-01-28 cs.CV

Shujuan Huang , Chunyu Lin , Yao Zhao This is my paper · ORCID
classification cs.CV
keywords LiDAR-cameracalibrationextrinsicregression-basednetworksretrievaldepthmapdistributioncross-modalmatchinggeneralizationdatagenerationpipeline
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 mainstream regression-based LiDAR-camera calibration networks, which are usually described as aligning image and point cloud features, actually learn a simpler thing: a lookup from the spatial distribution of the projected depth map to the extrinsic parameters. The authors build a minimal dual-branch network and probe it by testing on the right camera after training on the left. The x-translation error jumps to roughly the stereo baseline, the RGB branch can be removed with almost no loss, and relabeling the classification target with the left-to-right transform restores accuracy, all consistent with the network memorizing the training sensor's mapping rather than doing cross-modal matching. If correct, these methods cannot generalize to unseen sensor configurations, and the commonly used data-generation pipeline of random perturbations only teaches the network to undo perturbations for one fixed setup. The paper positions matching-based methods with explicit geometric solvers as the more principled path forward.

What carries the argument

The object that carries the argument is the projected depth map $D$ obtained by projecting LiDAR points through the camera intrinsics, and the regression mapping $F(D)\to T$ it induces. Because the extrinsic perturbation directly changes where points land in the image, similar perturbations create similar depth-map silhouettes, so a network can learn to read off the perturbation from the depth distribution alone, without ever relating image content to geometry. The paper's probe is the cross-camera test: since the left and right cameras share the same LiDAR but differ by a known stereo baseline, a genuine matcher trained on the left should transfer to the right, whereas a depth-distribution memorizer will predict the left camera's extrinsics and fail by the baseline offset. A second instrument is the label-relabeling experiment, where classification targets are recomputed from the left-to-right transform, which isolates whether the network stores a camera-specific mapping.

What would settle it

Train a regression-based calibrator on the left camera with the standard pipeline, then test on the right camera: if the x-translation error does not jump to roughly the stereo baseline, the retrieval claim is wrong. A sharper test is to train the same architecture on data synthesized with randomized camera intrinsics while keeping extrinsics fixed; if the network still generalizes across intrinsics, it has not simply memorized a depth-map-to-pose lookup.

Watch

Extended reading notes

Core claim

The central discovery is that the effective training objective of regression-based calibrators is $F(D)\to T$: a mapping from the projected depth map alone to the extrinsic transform, not $F(P,I)\to T$ from the joint point cloud and image. Evidence includes a cross-camera test where training on the left LiDAR-camera pair and testing on the right gives a translation error close to the stereo baseline while rotation degrades only mildly, ablating the RGB branch leaves accuracy nearly unchanged, and converting the regression head into a binary classifier shows the same x-translation drop, which is fully recovered when labels are recomputed using the left-to-right camera transform. The authors conclude that the networks cluster depth-map patterns and memorize the corresponding perturbations, so they never learn geometric correspondence. They also show the standard data-generation pipeline, $T_{gt}=T_{init}(\Delta T)^{-1}$, produces de-calibrated versions of one fixed setup rather than varied sensor configurations, and that preprocessing changes such as random cropping, densification, and point-cloud downsampling break performance by shifting the depth-map distribution the network has memorized.

Load-bearing premise

The argument hinges on assuming the left and right cameras on the test vehicle are similar enough that a network which truly matched images to point clouds would work for both; if those cameras differ in ways that change the depth-map appearance, part of the failure could be an input-statistics effect rather than proof that no matching is learned.

Editorial extensions

If this is right

  • Regression-based methods, including those that insert cost volumes or cross-attention modules, are still governed by the depth-map distribution and will not generalize to unseen LiDAR-camera configurations.
  • Reported accuracy on a fixed dataset can be high even when the network performs no actual cross-modal matching, so benchmark numbers for regression calibrators should be read with caution.
  • The standard perturbation-based data generation only creates de-calibrated samples of one setup, so it cannot teach a network to handle different sensor extrinsics or intrinsics.
  • Matching-based methods that predict correspondences and solve extrinsics with an explicit geometric solver retain the generalizability that regression-based methods lack.
  • Preprocessing choices such as cropping, resizing, densification, and sampling change the depth-map distribution and can break a regression-based calibrator, meaning practical deployments must either canonicalize inputs or include those variations in training.

Reading between the lines

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

  • If the retrieval interpretation is right, the same failure should appear in other regression-based sensor-calibration settings, such as radar-camera or camera-camera extrinsic calibration, and a cross-sensor probe analogous to the left-right test could expose it.
  • A testable extension is to train a regression calibrator on a large set of simulated sensor rigs with randomized extrinsics and intrinsics; if it still fails to extrapolate to a held-out rig, the paper's case that the regression paradigm itself is the bottleneck would be strengthened.
  • The depth-map-distribution clue suggests a simple diagnostic: perturb only the image branch by occluding, blurring, or shifting the RGB input and measure the prediction change; a matching-based network should react, while a retrieval network should not.
  • The label-relabeling trick could be turned into a data-augmentation strategy, generating training pairs from different virtual camera placements to force the network to encode the relative transform, potentially pushing regression-based methods closer to genuine matching.
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 / 5 minor

Summary. The paper investigates the working principles of learning-based LiDAR-camera calibration, focusing on regression-based methods. The authors argue that, despite being designed to perform cross-modal feature matching, regression-based networks actually learn a mapping from the projected depth map distribution to the extrinsic parameters, effectively acting as retrieval networks that memorize the training sensor setup. Evidence includes a cross-camera generalization test (training on KITTI left camera, testing on right camera), a depth-only vs. RGB-D ablation, a classification relabeling experiment, and analyses of the standard data generation pipeline. The paper concludes that the widely used perturbation-based data generation causes overfitting to the sensor configuration and that matching-based methods with explicit geometry are more promising for real-world generalization.

Significance. If the central claim holds, this is an important negative result for the community: it explains why regression-based LiDAR-camera calibration methods fail to generalize to new sensor setups and why they do not perform the cross-modal matching they are often described as doing. The paper contributes a systematic diagnostic methodology and a set of controlled experiments, which is valuable for guiding future research toward matching-based approaches. The paper is also useful in highlighting the limitations of the common perturbation-based training data generation. The evidence is partly convergent (depth-only ablation, classification relabeling), but the cross-camera test has a confounding factor, making the strongest interpretation depend on additional analysis.

major comments (3)
  1. [Section III-B, Table I] The cross-camera experiment is load-bearing for the retrieval claim, but it does not control for the differences in intrinsic parameters and field of view between the KITTI left and right cameras. The observed right-camera translation error jump could be explained by the network's sensitivity to the resulting depth-map distribution shift rather than by memorization of the left camera's extrinsics. To support Eq. (6), the paper should add an experiment that isolates the extrinsic shift from the intrinsic/FoV shift, for example by testing on the left camera with synthetic preprocessing that mimics the right camera's intrinsics, or by quantifying the residual error after correcting the predicted translation by the left-to-right transform. Without such a control, the cross-camera result alone does not distinguish the retrieval interpretation from a depth-map-to-pose function that is merely sensitive to camera intrinsics.
  2. [Section III-B, Fig. 4] The classification relabeling experiment is potentially the strongest evidence against the intrinsic-sensitivity alternative, but the paper does not explicitly connect it to the cross-camera confound. The authors should provide a quantitative analysis showing that the systematic prediction error equals the left-to-right transformation and that the residual error after applying this correction is small. This would demonstrate that the failure is driven by the memorized extrinsic mapping rather than by intrinsic/FoV differences. Currently, the relabeling experiment is presented as an 'alternative viewpoint' without being framed as a control for the cross-camera test.
  3. [Section II and Section III-B] The term 'retrieval' is used to characterize the network's behavior, but it is never precisely defined. The experiments show that the network generalizes across scenes on the same camera, which indicates it learns a smooth depth-map-to-pose function rather than a nearest-neighbor lookup of training samples. The paper should either define retrieval in a falsifiable way (e.g., via nearest-neighbor analysis in feature space) or soften the claim to state that regression-based methods learn a depth-map-to-pose mapping that is independent of RGB content and overfits to the training sensor configuration. As written, the retrieval terminology may overstate the degree of memorization.
minor comments (5)
  1. [Table I and Section III-B] The text says 'Matching-based methods such as CMRNet and MDPCalib show significantly better generalizability,' but CMRNet is elsewhere classified as a regression-based method; the intended reference is likely CMRNext. Please correct this inconsistency.
  2. [Section III-B, Fig. 4 caption] The 'transformation from the left camera to the right camera' is not defined; specify the exact matrix and its direction (e.g., T_LR such that T_R = T_LR * T_L) to make the relabeling procedure unambiguous.
  3. [Section III-A] The paper states that the simple framework follows CalibNet but does not provide sufficient architectural and training details (e.g., fully connected layer dimensions, input resolution, number of training iterations). These details are necessary for reproducibility.
  4. [Section III-B, classification experiment] The classification task is described only in one paragraph; please provide details on the number of classes, the loss function, and how the binary labels for each component are generated from the ground truth and the perturbation range.
  5. [Section III-C] When describing random cropping, the paper notes that 'intrinsic parameters should change when operations like cropping, padding and resizing are performed,' but it does not specify whether the depth maps are reprojected with adjusted intrinsics or simply cropped in image space. Clarify this procedure to support the interpretation in Table II.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the retrieval-network claim rests on external benchmark probes and ablations, not on restating its own inputs.

full rationale

The paper's central claim is that regression-based calibrators learn a depth-map-to-extrinsic mapping F(D)->T (Eq. 6, Sec. III-B) rather than performing cross-modal matching, and that this explains poor cross-camera generalization. This claim is supported by independent empirical probes: cross-camera tests on KITTI (Table I, Fig. 3), removal of the RGB branch (Table I), and the classification relabeling experiment (Fig. 4). These experiments are measured on external KITTI/KITTI-360 data and their outcomes are not derived from the conclusion. The simplification from F(D,I) to F(D) in Eq. 6 follows from the observed RGB ablation result, not from the definition of the loss or from a self-citation. The relabeling experiment is a genuine probe of whether predictions track the left-to-right camera transform; its outcome is not logically forced by the training objective. The assumption that the KITTI left and right cameras are similar enough for a matching-based calibrator to transfer is a substantive empirical premise, and its possible violation due to intrinsic or field-of-view differences is a correctness and robustness concern, not a circularity. There are no load-bearing self-citations: the cited works are prior methods and datasets from other research groups, not the authors' own prior results invoked to justify the central claim. Because no step in the derivation reduces by construction to its own inputs, the paper has no significant circularity.

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

The paper introduces no new entities or fitted constants. It relies on standard pinhole projection, plus domain assumptions about camera interchangeability and what random-perturbation data generation can teach a network. These assumptions are load-bearing for the retrieval claim.

assumptions (3)
  • domain assumption A network that performs cross-modal matching should be robust to swapping between two cameras with similar intrinsics and orientation.
    Used to interpret the right-camera performance drop as evidence of retrieval rather than sensitivity to intrinsics (Sec. III.B, Tab. I).
  • domain assumption Random perturbations applied to a fixed LiDAR-camera transform do not simulate realistic variation of sensor layouts.
    Used to argue the standard data generation pipeline is fundamentally limited (Sec. III.C).
  • domain assumption The classification task with labels regenerated using the identity of the test camera reveals what the network memorized.
    The classification experiment is interpreted as direct evidence of SE(3) mapping memorization (Sec. III.B, Fig. 4).

how reviews work

0 comments
Cite this review

Pith. "Pith review of What Really Matters for Learning-based LiDAR-Camera Calibration." pith.science (2026). https://pith.science/paper/Q4JEFBRH

@misc{pith2026250116969,
  author       = {Pith},
  title        = {Pith review of: What Really Matters for Learning-based LiDAR-Camera Calibration},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Q4JEFBRH}},
  note         = {Machine review of arXiv:2501.16969}
}
read the original abstract

Calibration is an essential prerequisite for the accurate data fusion of LiDAR and camera sensors. Traditional calibration techniques often require specific targets or suitable scenes to obtain reliable 2D-3D correspondences. To tackle the challenge of target-less and online calibration, deep neural networks have been introduced to solve the problem in a data-driven manner. While previous learning-based methods have achieved impressive performance on specific datasets, they still struggle in complex real-world scenarios. Most existing works focus on improving calibration accuracy but overlook the underlying mechanisms. In this paper, we revisit the development of learning-based LiDAR-Camera calibration and encourage the community to pay more attention to the underlying principles to advance practical applications. We systematically analyze the paradigm of mainstream learning-based methods, and identify the critical limitations of regression-based methods with the widely used data generation pipeline. Our findings reveal that most learning-based methods inadvertently operate as retrieval networks, focusing more on single-modality distributions rather than cross-modality correspondences. We also investigate how the input data format and preprocessing operations impact network performance and summarize the regression clues to inform further improvements.

Figures

Figures reproduced from arXiv: 2501.16969 by the authors.

Figure 1
Figure 1. The transformation between a point cloud captured from LiDAR [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Two mainstream learning-based LiDAR-Camera calibration framework, from the perspective of the output format of the network. Regression-based [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Rotation and translation error distributions of the cross-camera [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (1 more)
Figure 6
Figure 6. Figure 6: Illustration of the performance drop in cross-camera generalization [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

66 extracted references · 58 canonical work pages

  1. [1]

    Fast-livo: Fast and tightly-coupled sparse-direct lidar- inertial-visual odometry

    Chunran Zheng, Qingyan Zhu, Wei Xu, Xiyuan Liu, Qizhi Guo, and Fu Zhang. Fast-livo: Fast and tightly-coupled sparse-direct lidar- inertial-visual odometry. In 2022 IEEE/RSJ international conference on intelligent robots and systems (IROS) , pages 4003–4009. IEEE, 2022

    work page 2022

  2. [2]

    Bevfusion: Multi-task multi-sensor fusion with unified bird’s-eye view representation

    Zhijian Liu, Haotian Tang, Alexander Amini, Xinyu Yang, Huizi Mao, Daniela L Rus, and Song Han. Bevfusion: Multi-task multi-sensor fusion with unified bird’s-eye view representation. In 2023 IEEE international conference on robotics and automation (ICRA) , pages 2774–2781. IEEE, 2023

    work page 2023

  3. [3]

    Smart roads for autonomous accident detection and warnings

    Abdul Mateen, Muhammad Zahid Hanif, Narayan Khatri, Sihyung Lee, and Seung Yeob Nam. Smart roads for autonomous accident detection and warnings. Sensors, 22(6):2077, 2022

    work page 2022

  4. [4]

    Transfusion: Robust lidar-camera fusion for 3d object detection with transformers

    Xuyang Bai, Zeyu Hu, Xinge Zhu, Qingqiu Huang, Yilun Chen, Hongbo Fu, and Chiew-Lan Tai. Transfusion: Robust lidar-camera fusion for 3d object detection with transformers. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 1090–1099, 2022

    2022

  5. [5]

    Uniseg: A unified multi-modal lidar segmentation network and the openpcseg codebase

    Youquan Liu, Runnan Chen, Xin Li, Lingdong Kong, Yuchen Yang, Zhaoyang Xia, Yeqi Bai, Xinge Zhu, Yuexin Ma, Yikang Li, et al. Uniseg: A unified multi-modal lidar segmentation network and the openpcseg codebase. In Proceedings of the IEEE/CVF International Conference on Computer Vision , pages 21662–21673, 2023

    work page 2023

  6. [6]

    Rail vehicle localization and mapping with lidar-vision-inertial-gnss fusion

    Yusheng Wang, Weiwei Song, Yidong Lou, Yi Zhang, Fei Huang, Zhiyong Tu, and Qiangsheng Liang. Rail vehicle localization and mapping with lidar-vision-inertial-gnss fusion. IEEE Robotics and Automation Letters, 7(4):9818–9825, 2022

    work page 2022

  7. [7]

    Openoc- cupancy: A large scale benchmark for surrounding semantic occupancy perception

    Xiaofeng Wang, Zheng Zhu, Wenbo Xu, Yunpeng Zhang, Yi Wei, Xu Chi, Yun Ye, Dalong Du, Jiwen Lu, and Xingang Wang. Openoc- cupancy: A large scale benchmark for surrounding semantic occupancy perception. In Proceedings of the IEEE/CVF International Conference on Computer Vision , pages 17850–17859, 2023

    work page 2023

  8. [8]

    A flexible new technique for camera calibration

    Zhengyou Zhang. A flexible new technique for camera calibration. IEEE Transactions on pattern analysis and machine intelligence , 22(11):1330–1334, 2000

    work page 2000

Show all 66 references
  1. [9]

    Line-based extrinsic calibration of range and image sensors

    Peyman Moghadam, Michael Bosse, and Robert Zlot. Line-based extrinsic calibration of range and image sensors. In 2013 IEEE International Conference on Robotics and Automation , pages 3685–

    2013

  2. [10]

    Automatic online calibration of cameras and lasers

    Jesse Levinson and Sebastian Thrun. Automatic online calibration of cameras and lasers. In Robotics: science and systems , volume 2. Citeseer, 2013

    2013

  3. [11]

    Sfm and semantic information based online targetless camera-lidar self-calibration

    Bal ´azs Nagy, Levente Kov´acs, and Csaba Benedek. Sfm and semantic information based online targetless camera-lidar self-calibration. In 2019 IEEE International Conference on Image Processing (ICIP) , pages 1317–1321. IEEE, 2019

    2019

  4. [12]

    Crlf: Automatic calibration and refinement based on line feature for lidar and camera in road scenes

    Tao Ma, Zhizheng Liu, Guohang Yan, and Yikang Li. Crlf: Automatic calibration and refinement based on line feature for lidar and camera in road scenes. arXiv preprint arXiv:2103.04558 , 2021

    2021 arXiv

  5. [13]

    Regnet: Multimodal sensor registration using deep neural networks

    Nick Schneider, Florian Piewak, Christoph Stiller, and Uwe Franke. Regnet: Multimodal sensor registration using deep neural networks. In 2017 IEEE intelligent vehicles symposium (IV) , pages 1803–1810. IEEE, 2017

    2017

  6. [14]

    Calibnet: Geometrically supervised extrinsic calibration using 3d spatial transformer networks

    Ganesh Iyer, R Karnik Ram, J Krishna Murthy, and K Madhava Krishna. Calibnet: Geometrically supervised extrinsic calibration using 3d spatial transformer networks. In 2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pages 1110–

    2018

  7. [15]

    Cmrnet: Cam- era to lidar-map registration

    Daniele Cattaneo, Matteo Vaghi, Augusto Luis Ballardini, Simone Fontana, Domenico G Sorrenti, and Wolfram Burgard. Cmrnet: Cam- era to lidar-map registration. In 2019 IEEE intelligent transportation systems conference (ITSC) , pages 1283–1289. IEEE, 2019

    2019

  8. [16]

    Extrinsic calibration of a camera and laser range finder (improves camera calibration)

    Qilong Zhang and Robert Pless. Extrinsic calibration of a camera and laser range finder (improves camera calibration). In 2004 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)(IEEE Cat. No. 04CH37566) , volume 3, pages 2301–2306. IEEE, 2004

    2004

  9. [17]

    Fast extrinsic calibration of a laser rangefinder to a camera

    Ranjith Unnikrishnan and Martial Hebert. Fast extrinsic calibration of a laser rangefinder to a camera. Robotics Institute, Pittsburgh, PA, Tech. Rep. CMU-RI-TR-05-09, 2005

    2005

  10. [18]

    Extrinsic calibration of a 3d laser scanner and an omnidirectional camera

    Gaurav Pandey, James McBride, Silvio Savarese, and Ryan Eustice. Extrinsic calibration of a 3d laser scanner and an omnidirectional camera. IFAC Proceedings Volumes, 43(16):336–341, 2010

    2010

  11. [19]

    Joint camera intrinsic and lidar-camera extrinsic calibration

    Guohang Yan, Feiyu He, Chunlei Shi, Pengjin Wei, Xinyu Cai, and Yikang Li. Joint camera intrinsic and lidar-camera extrinsic calibration. In 2023 IEEE International Conference on Robotics and Automation (ICRA), pages 11446–11452. IEEE, 2023

    2023

  12. [20]

    Robust extrinsic self-calibration of camera and solid state lidar

    Jiahui Liu, Xingqun Zhan, Cheng Chi, Xin Zhang, and Chuanrun Zhai. Robust extrinsic self-calibration of camera and solid state lidar. Journal of Intelligent & Robotic Systems , 109(4):81, 2023

    2023

  13. [21]

    Automated extrinsic laser and camera inter-calibration using triangular targets

    Stefano Debattisti, Luca Mazzei, and Matteo Panciroli. Automated extrinsic laser and camera inter-calibration using triangular targets. In 2013 IEEE Intelligent Vehicles Symposium (IV), pages 696–701. IEEE, 2013

    2013

  14. [22]

    Calibration of camera and flash lidar system with a triangular pyramid target

    Zean Bu, Changku Sun, Peng Wang, and Hang Dong. Calibration of camera and flash lidar system with a triangular pyramid target. Applied sciences, 11(2):582, 2021

    2021

  15. [23]

    Extrinsic calibration of lidar and camera with polygon

    Qinghai Liao, Zhenyong Chen, Yang Liu, Zhe Wang, and Ming Liu. Extrinsic calibration of lidar and camera with polygon. In 2018 IEEE International Conference on Robotics and Biomimetics (ROBIO), pages 200–205. IEEE, 2018

    2018

  16. [24]

    Automatic extrinsic parameter calibration for camera-lidar fusion using spherical target

    Guanyu Zhang, Kunyang Wu, Jun Lin, Tianhao Wang, and Yang Liu. Automatic extrinsic parameter calibration for camera-lidar fusion using spherical target. IEEE Robotics and Automation Letters , 2024

    2024

  17. [25]

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

    Ankit Dhall, Kunal Chelani, Vishnu Radhakrishnan, and K Madhava Krishna. Lidar-camera calibration using 3d-3d point correspondences. arXiv preprint arXiv:1705.09785 , 2017

    2017 arXiv

  18. [26]

    Automatic camera and range sensor calibration using a single shot

    Andreas Geiger, Frank Moosmann, ¨Omer Car, and Bernhard Schuster. Automatic camera and range sensor calibration using a single shot. In 2012 IEEE international conference on robotics and automation , pages 3936–3943. IEEE, 2012

    2012

  19. [27]

    Omnidirectional field of view structured light calibration method for catadioptric vision system

    Xin Chen, Fuqiang Zhou, and Ting Xue. Omnidirectional field of view structured light calibration method for catadioptric vision system. Measurement, 148:106914, 2019

    2019

  20. [28]

    Omnidirectional vision system with laser illumination in a flexible configuration and its cali- bration by one single snapshot

    Ivan Kholodilin, Yuan Li, and Qinglin Wang. Omnidirectional vision system with laser illumination in a flexible configuration and its cali- bration by one single snapshot. IEEE Transactions on Instrumentation and Measurement, 69(11):9105–9118, 2020

    2020

  21. [29]

    Au- tocalibration of lidar and optical cameras via edge alignment

    Juan Castorena, Ulugbek S Kamilov, and Petros T Boufounos. Au- tocalibration of lidar and optical cameras via edge alignment. In 2016 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 2862–2866. IEEE, 2016

    2016

  22. [30]

    Lidar-camera calibration using line correspondences

    Zixuan Bai, Guang Jiang, and Ailing Xu. Lidar-camera calibration using line correspondences. Sensors, 20(21):6319, 2020

    2020

  23. [31]

    Camvox: A low-cost and accurate lidar-assisted visual slam system

    Yuewen Zhu, Chunran Zheng, Chongjian Yuan, Xu Huang, and Xiaoping Hong. Camvox: A low-cost and accurate lidar-assisted visual slam system. In 2021 IEEE International Conference on Robotics and Automation (ICRA), pages 5049–5055. IEEE, 2021

    2021

  24. [32]

    Pixel- level extrinsic self calibration of high resolution lidar and camera in targetless environments

    Chongjian Yuan, Xiyuan Liu, Xiaoping Hong, and Fu Zhang. Pixel- level extrinsic self calibration of high resolution lidar and camera in targetless environments. IEEE Robotics and Automation Letters , 6(4):7517–7524, 2021

    2021

  25. [33]

    The road is enough! extrinsic calibration of non-overlapping stereo camera and lidar using road information

    Jinyong Jeong, Younghun Cho, and Ayoung Kim. The road is enough! extrinsic calibration of non-overlapping stereo camera and lidar using road information. IEEE Robotics and Automation Letters , 4(3):2831– 2838, 2019

    2019

  26. [34]

    On-the-fly camera and lidar calibration

    Bal ´azs Nagy and Csaba Benedek. On-the-fly camera and lidar calibration. Remote Sensing, 12(7):1137, 2020

    2020

  27. [35]

    From chaos to calibration: A geometric mutual information approach to target-free camera lidar extrinsic calibration

    Jack Borer, Jeremy Tschirner, Florian ¨Olsner, and Stefan Milz. From chaos to calibration: A geometric mutual information approach to target-free camera lidar extrinsic calibration. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision , pages 840...

    2024

  28. [36]

    Automatic targetless extrinsic calibration of a 3d lidar and camera by maximizing mutual information

    Gaurav Pandey, James McBride, Silvio Savarese, and Ryan Eustice. Automatic targetless extrinsic calibration of a 3d lidar and camera by maximizing mutual information. In Proceedings of the AAAI conference on artificial intelligence , volume 26, pages 2053–2059, 2012

    2012

  29. [37]

    Automatic calibration of lidar and camera images using normalized mutual information

    Zachary Taylor and Juan Nieto. Automatic calibration of lidar and camera images using normalized mutual information. In Robotics and Automation (ICRA), 2013 IEEE International Conference on. Citeseer, 2013

    2013

  30. [38]

    Motion-based calibration of multi- modal sensor extrinsics and timing offset estimation

    Zachary Taylor and Juan Nieto. Motion-based calibration of multi- modal sensor extrinsics and timing offset estimation. IEEE Transac- tions on Robotics , 32(5):1215–1229, 2016

    2016

  31. [39]

    Lidar and camera calibration using motions estimated by sensor fusion odometry

    Ryoichi Ishikawa, Takeshi Oishi, and Katsushi Ikeuchi. Lidar and camera calibration using motions estimated by sensor fusion odometry. In 2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 7342–7349. IEEE, 2018

    2018

  32. [40]

    Cfnet: Lidar-camera registra- tion using calibration flow network

    Xudong Lv, Shuo Wang, and Dong Ye. Cfnet: Lidar-camera registra- tion using calibration flow network. Sensors, 21(23):8112, 2021

    2021

  33. [41]

    Ep n p: An accurate o (n) solution to the p n p problem

    Vincent Lepetit, Francesc Moreno-Noguer, and Pascal Fua. Ep n p: An accurate o (n) solution to the p n p problem. International journal of computer vision , 81:155–166, 2009

    2009

  34. [42]

    Dxq-net: Differentiable lidar-camera extrinsic calibration using quality-aware flow

    Xin Jing, Xiaqing Ding, Rong Xiong, Huanjun Deng, and Yue Wang. Dxq-net: Differentiable lidar-camera extrinsic calibration using quality-aware flow. In 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 6235–6241. IEEE, 2022

    2022

  35. [43]

    Cmrnet++: Map and camera agnostic monocular visual localization in lidar maps

    D Sorrenti, C Daniele, A Valada, et al. Cmrnet++: Map and camera agnostic monocular visual localization in lidar maps. In Proceed- ing of ICRA 2020 Workshop on Emerging Learning and Algorith- mic Methods for Data Association in Robotics https://sites. google. com/view/edat/hom...

    2020 arXiv

  36. [44]

    Cmrnext: Camera to lidar matching in the wild for localization and extrinsic calibration

    Daniele Cattaneo and Abhinav Valada. Cmrnext: Camera to lidar matching in the wild for localization and extrinsic calibration. arXiv preprint arXiv:2402.00129, 2024

    2024 arXiv

  37. [45]

    Automatic target-less camera- lidar calibration from motion and deep point correspondences

    K ¨ursat Petek, Niclas V ¨odisch, Johannes Meyer, Daniele Cattaneo, Abhinav Valada, and Wolfram Burgard. Automatic target-less camera- lidar calibration from motion and deep point correspondences. arXiv preprint arXiv:2404.17298, 2024

    2024 arXiv

  38. [46]

    Keypoint-based lidar-camera online calibration with robust geometric network

    Chao Ye, Huihui Pan, and Huijun Gao. Keypoint-based lidar-camera online calibration with robust geometric network. IEEE Transactions on Instrumentation and Measurement , 71:1–11, 2021

    2021

  39. [47]

    Atop: An attention-to-optimization approach for automatic lidar- camera calibration via cross-modal object matching

    Yi Sun, Jian Li, Yuru Wang, Xin Xu, Xiaohui Yang, and Zhenping Sun. Atop: An attention-to-optimization approach for automatic lidar- camera calibration via cross-modal object matching. IEEE Transac- tions on Intelligent Vehicles , 8(1):696–708, 2022

    2022

  40. [48]

    Rggnet: Tolerance aware lidar-camera online calibration with geometric deep learning and generative model

    Kaiwen Yuan, Zhenyu Guo, and Z Jane Wang. Rggnet: Tolerance aware lidar-camera online calibration with geometric deep learning and generative model. IEEE Robotics and Automation Letters , 5(4):6956– 6963, 2020

    2020

  41. [49]

    A robust lidar-camera self-calibration via rotation-based alignment and multi-level cost volume

    Zaipeng Duan, Xuzhong Hu, Junfeng Ding, Pei An, Xiao Huang, and Jie Ma. A robust lidar-camera self-calibration via rotation-based alignment and multi-level cost volume. IEEE Robotics and Automation Letters, 9(1):627–634, 2023

    2023

  42. [50]

    Net- calib: A novel approach for lidar-camera auto-calibration based on deep learning

    Shan Wu, Amnir Hadachi, Damien Vivet, and Yadu Prabhakar. Net- calib: A novel approach for lidar-camera auto-calibration based on deep learning. In 2020 25th International Conference on Pattern Recognition (ICPR), pages 6648–6655. IEEE, 2021

    2020

  43. [51]

    Calibdnn: multimodal sensor calibration for perception using deep neural net- works

    Ganning Zhao, Jiesi Hu, Suya You, and C-C Jay Kuo. Calibdnn: multimodal sensor calibration for perception using deep neural net- works. In Signal Processing, Sensor/Information Fusion, and Target Recognition XXX, volume 11756, pages 324–335. SPIE, 2021

    2021

  44. [52]

    Psnet: Lidar and camera registration using parallel subnetworks

    Yi Wu, Ming Zhu, and Ji Liang. Psnet: Lidar and camera registration using parallel subnetworks. IEEE Access, 10:70553–70561, 2022

    2022

  45. [53]

    Mrcnet: Multi-resolution lidar-camera calibration using optical center distance loss network

    Hao Wang, Zhangyu Wang, Guizhen Yu, Songyue Yang, and Yang Yang. Mrcnet: Multi-resolution lidar-camera calibration using optical center distance loss network. IEEE Sensors Journal , 2023

    2023

  46. [54]

    Calibrcnn: Calibrating camera and lidar by recurrent convolutional neural network and geometric constraints

    Jieying Shi, Ziheng Zhu, Jianhua Zhang, Ruyu Liu, Zhenhua Wang, Shengyong Chen, and Honghai Liu. Calibrcnn: Calibrating camera and lidar by recurrent convolutional neural network and geometric constraints. In 2020 IEEE/RSJ International Conference on Intelligent Robots and Sys...

    2020

  47. [55]

    Calibbd: Extrinsic calibration of the lidar and camera using a bidirectional neural network

    An Duy Nguyen and Myungsik Yoo. Calibbd: Extrinsic calibration of the lidar and camera using a bidirectional neural network. IEEE Access, 10:121261–121271, 2022

    2022

  48. [56]

    Calnet: Lidar-camera online calibration with channel attention and liquid time-constant network

    Hongcheng Shang and Bin-Jie Hu. Calnet: Lidar-camera online calibration with channel attention and liquid time-constant network. In 2022 26th International Conference on Pattern Recognition (ICPR) , pages 5147–5154. IEEE, 2022

    2022

  49. [57]

    Robust lidar- camera alignment with modality adapted local-to-global representa- tion

    Angfan Zhu, Yang Xiao, Chengxin Liu, and Zhiguo Cao. Robust lidar- camera alignment with modality adapted local-to-global representa- tion. IEEE Transactions on Circuits and Systems for Video Technology, 33(1):59–73, 2022

    2022

  50. [58]

    Calibdepth: Unifying depth map representation for iterative lidar-camera online calibration

    Jiangtong Zhu, Jianru Xue, and Pu Zhang. Calibdepth: Unifying depth map representation for iterative lidar-camera online calibration. In 2023 IEEE International Conference on Robotics and Automation (ICRA), pages 726–733. IEEE, 2023

    2023

  51. [59]

    Dedgenet: Extrinsic calibration of camera and lidar with depth-discontinuous edges

    Yiyang Hu, Hui Ma, Leiping Jie, and Hui Zhang. Dedgenet: Extrinsic calibration of camera and lidar with depth-discontinuous edges. In 2023 IEEE International Conference on Robotics and Automation (ICRA), pages 11439–11445. IEEE, 2023

    2023

  52. [60]

    Lccnet: Lidar and camera self-calibration using cost volume network

    Xudong Lv, Boya Wang, Ziwen Dou, Dong Ye, and Shuo Wang. Lccnet: Lidar and camera self-calibration using cost volume network. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 2894–2901, 2021

    2021

  53. [61]

    Fusionnet: Coarse-to-fine extrinsic calibration network of lidar and camera with hierarchical point-pixel fusion

    Guangming Wang, Jiahao Qiu, Yanfeng Guo, and Hesheng Wang. Fusionnet: Coarse-to-fine extrinsic calibration network of lidar and camera with hierarchical point-pixel fusion. In 2022 International Conference on Robotics and Automation (ICRA) , pages 8964–8970. IEEE, 2022

    2022

  54. [62]

    Robust online calibration of lidar and camera based on cross-modal graph neural network

    Jianxiao Zhu, Xu Li, Qimin Xu, and Zhengliang Sun. Robust online calibration of lidar and camera based on cross-modal graph neural network. IEEE Transactions on Instrumentation and Measurement , 2023

    2023

  55. [63]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition , pages 770– 778, 2016

    2016

  56. [64]

    Are we ready for autonomous driving? the kitti vision benchmark suite

    Andreas Geiger, Philip Lenz, and Raquel Urtasun. Are we ready for autonomous driving? the kitti vision benchmark suite. In Conference on Computer Vision and Pattern Recognition (CVPR) , 2012

    2012

  57. [65]

    Kitti-360: A novel dataset and benchmarks for urban scene understanding in 2d and 3d

    Yiyi Liao, Jun Xie, and Andreas Geiger. Kitti-360: A novel dataset and benchmarks for urban scene understanding in 2d and 3d. IEEE Trans- actions on Pattern Analysis and Machine Intelligence , 45(3):3292– 3310, 2022

    2022

  58. [66]

    Adam: A method for stochastic optimization

    Diederik P Kingma. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 , 2014

    2014 arXiv

Pith tools

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