Pith. sign in

REVIEW 4 major objections 3 minor 23 references

SuperMapNet for Long-Range and High-Accuracy Vectorized HD Map Construction

T0 review · 4 major / 3 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read SuperMapNet couples camera semantics with LiDAR geometry and point-level with element-level modeling to construct vectorized HD maps over a 120 m range, reporting state-of-the-art mAP on nuScenes and Argoverse2.

desk verdict Sensible architecture and honest ablations, but the headline SOTA margins rest on cross-protocol comparisons and do not hold up as printed. read the letter →

arxiv 2505.13856 v2 pith:5JPMCEEF submitted 2025-05-20 cs.CV

Ruqin Zhou , Chenguang Dai , Wanshou Jiang , Yongsheng Zhang , Hanyun Wang , San Jiang This is my paper · ORCID
classification cs.CV
keywords vectorizedHDmapmulti-modalfusionbird's-eye-viewLiDARpointcloudscross-attentionpoint-elementcouplinglong-rangeperceptionautonomousdriving
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

SuperMapNet's claim is that vectorized HD map construction can be pushed to a 120 m perception range and to higher accuracy by treating fusion and decoding as structured coupling problems rather than concatenation. On the fusion side, it couples camera semantics with LiDAR geometry using cross-attention synergy enhancement plus flow-based disparity alignment. On the decoding side, it couples point-level and element-level modeling through three interactions: Point2Point, Element2Element, and Point2Element. The reported results on nuScenes and Argoverse2, new best mAP under both hard and easy thresholds, would matter because autonomous driving needs maps that are both long-range and accurate at the centimeter-to-decimeter level.

What carries the argument

The load-bearing machinery is the pair of coupling modules. The SGC module has two sub-modules: a cross-attention synergy enhancement that treats camera BEV features and LiDAR BEV features as query/key/value pairs in both directions, so each modality's features absorb complementary information from the other; and a flow-based disparity alignment that predicts a per-cell displacement $(\Delta h, \Delta w)$ and bilinearly resamples the camera BEV features onto LiDAR coordinates before the final concatenation. The PEC module operates on ordered keypoint sets with learnable point queries and element queries, and uses three attention levels, Point2Point, Element2Element, and Point2Element, where Point2Element updates point descriptors with element descriptors and vice versa via cross-attention with position embedding. These modules implement the paper's two premises: fusion must manage cross-modal synergy and sensor disparity, and decoding must use both point-level and element-level information.

What would settle it

Run the strongest prior camera-plus-LiDAR method under SuperMapNet's own protocol, with a 120 m Y-axis range, 0.15 m voxelization, hard thresholds {0.2, 0.5, 1.0} m and easy thresholds {0.5, 1.0, 1.5} m, on the same splits, and compute mAP; if hard/easy mAP meets or exceeds 66.5/86.6 on nuScenes or 64.3/82.4 on Argoverse2, the SOTA claim collapses. Alternatively, replace SGC and PEC in SuperMapNet with plain concatenation and a point-only decoder under the same 120 m range; equal accuracy would falsify the mechanism claim.

Watch

Extended reading notes

Core claim

The central claim is that two specific design choices explain high-accuracy long-range vectorized HD mapping. First, the semantic-geometric coupling (SGC) module fuses camera and LiDAR BEV features not by direct concatenation but by cross-attention, letting each modality fill the other's gaps, followed by a learned flow-based alignment that registers camera features onto the more accurate LiDAR coordinates before concatenation. Second, the point-element coupling (PEC) module decodes map elements with three levels of interaction: Point2Point for intra-element geometry, Element2Element for inter-element semantic constraints, and Point2Element for passing global element context back to points. With both modules, SuperMapNet reports 66.5/86.6 mAP on nuScenes and 64.3/82.4 mAP on Argoverse2 under its hard/easy AP thresholds, corresponding to claimed gains of 14.9/8.8 and 18.5/3.1 mAP over the second-best compared methods, over a 120 m Y-axis range that is twice the prior 60 m.

Load-bearing premise

The reported state-of-the-art gains assume that the earlier results quoted for comparison used the same evaluation rules, same map coverage windows, same data splits, and same matching thresholds, so that the accuracy gaps compare like with like.

Editorial extensions

If this is right

  • If the 120 m operating point is adopted by later systems, downstream planning modules can receive map information twice as far ahead, changing when lane-change and braking decisions can be made.
  • The claimed gains imply that direct concatenation of camera and LiDAR BEV features leaves measurable accuracy unused, specifically about 4.9/3.2 mAP on nuScenes validation for SGC alone.
  • The PEC ablation suggests element-level context is worth more than the fusion module alone on this task, with +7.7/+6.7 mAP, and is especially valuable for elements modeled by many keypoints.
  • At the strict 0.2 m threshold, the combined SGC+PEC gain is +11.1 mAP, so the modules matter most where centimeter-level accuracy is demanded.

Reading between the lines

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

  • A natural extension the paper does not test is whether the flow-based disparity alignment is robust to deliberate miscalibration; injecting synthetic translation or rotation errors between camera and LiDAR during evaluation would show whether the module actually corrects sensor disparity or merely learns dataset-specific offsets.
  • Because the hard-setting second-best in Table 1 is camera-only, the hard-setting margin may shrink if the strongest multi-modal baseline publishes a hard-setting number; the easy setting is where a camera-plus-LiDAR comparison exists.
  • The degradation of SGC's benefit at thresholds above 0.5 m, if replicated, implies sensor alignment is a precision enabler rather than a general accuracy booster; systems targeting meter-level maps could skip disparity alignment and save compute.
  • Point2Point's weak effect on pedestrian crossings, which are modeled with only two keypoints, suggests a testable modification: allocate more keypoints or an alternative primitive for short elements.
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 / 3 minor

Summary. The paper proposes SuperMapNet, a camera-LiDAR framework for online vectorized HD map construction. It couples camera and LiDAR BEV features through a cross-attention synergy enhancement module and a flow-based disparity alignment module (SGC), and couples point queries with element queries through three interactions, Point2Point, Element2Element, and Point2Element (PEC). Experiments on nuScenes and Argoverse2 report 66.5/86.6 mAP and 64.3/82.4 mAP under the paper's hard/easy settings, which the authors interpret as 14.9/8.8 and 18.5/3.1 mAP gains over the second-best methods, at a perception range of 120 m in the Y-axis. The paper includes module-level ablations and threshold-sensitivity experiments.

Significance. If the reported numbers are taken at face value, the gains are large and the architecture is of genuine interest to the HD map construction community. The SGC module targets a real limitation of direct multi-modal concatenation, the PEC module explicitly models element-level relations that point-only methods ignore, the ablation study isolates the contribution of each module, and the code is publicly released. However, the central benchmark-superiority claim is not currently established because the comparisons mix evaluation protocols, and because the hard-setting gain is computed against a camera-only baseline. The architectural contribution remains plausible independently of the headline margins.

major comments (4)
  1. [§4.1, Tables 1 and 2] The headline SOTA margins are computed from numbers that are not demonstrably produced under a common protocol. Section 4.1 sets the perception range to [-15,15] m in X and [-60,60] m in Y and defines the hard/easy AP threshold sets {0.2,0.5,1.0} and {0.5,1.0,1.5}, while the table captions state that all comparative results are 'referenced in their paper.' The paper itself notes in Section 1 and Section 4.2 that other methods operate at roughly 60 m in Y, and in Section 4.1 that existing methods used different AP thresholds. If the cited baseline numbers were obtained at a 60 m range or under different AP thresholds, subtracting them from SuperMapNet's 120 m-range mAP is not a valid head-to-head comparison. Because the abstract and Section 4.2 state the gains as the central claim, the authors should either re-evaluate the baselines under the same range and threshold protocol or explicitly restrict the claim to a cross-protocol comparison and discuss the direction and magnitude of the resulting bias.
  2. [§4.2, Table 1] In the hard setting, the 'second-best' used to compute the +14.9 mAP gain is camera-only HIMapNet (51.6 mAP), because the multi-modal HIMapNet C+L row has no hard-setting entry ('-'). The gain statement does not disclose that the comparison is against a camera-only method rather than the strongest comparable multi-modal baseline. This weakens the hard-setting SOTA claim as printed. The authors should report a hard-setting result for HIMapNet C+L under the same protocol, or revise the claim and its wording to state explicitly which baseline the gain is measured against.
  3. [§4.2, Tables 1 and 2] The 'long-range' claim is not supported by any range-stratified evaluation. The paper reports aggregate mAP over a 120 m Y-range and repeatedly contrasts this with the 60 m range of prior methods, but Table 1 and Table 2 contain no per-range breakdown and no same-protocol baseline at 60 m. Without such data, the reader cannot tell how much of the reported margin comes from the SGC module's long-range fusion and how much from the different evaluation range. A complementary experiment evaluating SuperMapNet and at least one baseline at both 60 m and 120 m would make the long-range contribution testable.
  4. [§5.2, Table 4] The threshold ablation shows that the combination of SGC and PEC is substantially better than either module alone only at the strict 0.2 m threshold, while at 1.0 m and 1.5 m it matches PEC-only performance. This is an honest and useful observation, but the text in Section 5.2 that says the advantage 'basically does not exist, and achieves the same accuracy as only using PEC module, as the SGC module' is garbled: Table 4 shows SGC-only is lower at those thresholds, so the final phrase should be corrected to 'not as the SGC module' or reworded.
minor comments (3)
  1. [§3.2, Eqs. (10)-(12)] The disparity alignment equations contain an indexing/typographical error: Eq. (11) defines weight_w' = max(0, 1 - |h + Δh - h|), which reduces to |Δh| and cannot be the intended weight for the w-coordinate. The source and target coordinate indices (h, w, h', w', and any vertical-flow variable) should be defined consistently before Eqs. (10)-(12).
  2. [Table 2] In Table 2, the ADMap row is cited as 'ADMap (Wilson et al., 2021)', but ADMap is the method of Hu et al. (2024); the citation is incorrect and should be fixed.
  3. [§4.2] The statement that SuperMapNet 'maintains real-time capability' at 5.0 FPS is not self-evident for an online HD map construction system. If 5 FPS is considered real-time in this application context, the criterion should be stated explicitly; otherwise the claim should be softened to a latency report.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper is an empirical architecture study with internally controlled ablations; SOTA-margin caveats concern comparability, not derivation.

full rationale

This is an empirical systems paper rather than a derivation, and no load-bearing step reduces to its own inputs. The proposed SGC and PEC modules are defined by explicit network operations (cross-attention, disparity flow, point/element interactions), and the ablations in Tables 3 and 4 compare variants of the same model under a fixed protocol, so the module contributions are internally controlled. The headline SOTA claims are computed from numbers 'referenced in their paper' for prior methods, while SuperMapNet is evaluated at [-15,15] m in X and [-60,60] m in Y, and the paper acknowledges that 'existing methods used different AP thresholds for evaluation' before defining its own hard/easy threshold sets. This creates a legitimate concern about cross-protocol comparability of the reported margins, but it is not circularity: the measured mAPs of SuperMapNet are not constructed from the baseline numbers, no fitted parameter is renamed as a prediction, and no equation uses a target conclusion as an input. The self-defined thresholds and perception range affect what the comparison shows, but they do not make the reported accuracy values equivalent to the evaluation choice by construction. No self-citation chain is load-bearing, and no uniqueness theorem or ansatz is smuggled in from the authors' prior work. Therefore no circular step can be quoted, and the honest finding is no significant circularity.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central claim is an empirical benchmark result, so the ledger is dominated by evaluation choices and domain assumptions rather than mathematical axioms. The listed free parameters are hand-set hyperparameters and evaluation settings that directly shape the reported mAP; none are physical constants. No invented entities are introduced.

free parameters (5)
  • AP threshold sets for hard/easy evaluation = hard: {0.2, 0.5, 1.0} m; easy: {0.5, 1.0, 1.5} m
    Hand-chosen thresholds define the reported mAP and the headline gains. Prior methods used varying thresholds, making cross-paper comparability uncertain.
  • Perception range in Y-axis = [-60, 60] m (120 m total)
    Double the range of many compared methods, which changes task difficulty and makes the 'long-range' claim part of the evaluation setup.
  • Maximum number of elements per type M = lane dividers 20, pedestrian crossings 25, road boundaries 15
    Hand-set query counts affect capacity and matching; values are not derived from data.
  • Maximum keypoints per element N = lane dividers 10, pedestrian crossings 2, road boundaries 30
    Hand-set; pedestrian crossings are modeled by only two keypoints, which the authors note makes Point2Point interaction less effective.
  • BEV grid and voxel size = BEV 100x25, voxel 0.15 m
    Chosen resolution affects the precision ceiling and computational cost.
assumptions (4)
  • domain assumption The annotated ground truth in nuScenes and Argoverse2 is correct and consistent across categories.
    The entire evaluation and claimed SOTA depend on annotation quality; the paper does not audit labels.
  • domain assumption Sensor calibration between cameras and LiDAR is accurate enough that residual errors can be modeled by a small learned disparity flow.
    The SGC disparity alignment module (Eq. 9-12) presumes misalignment is small and can be corrected by bilinear resampling; large calibration errors would break fusion.
  • domain assumption Reported mAP numbers for compared methods in Tables 1 and 2 were computed under the same thresholds and evaluation code as used for SuperMapNet.
    The paper cites numbers from other papers directly; any protocol mismatch invalidates the SOTA comparison.
  • standard math Chamfer-distance-based average precision is a valid metric for HD map construction quality.
    The paper follows prior work in using AP with Chamfer distance, but does not discuss its sensitivity or alternatives.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SuperMapNet for Long-Range and High-Accuracy Vectorized HD Map Construction." pith.science (2026). https://pith.science/paper/5JPMCEEF

@misc{pith2026250513856,
  author       = {Pith},
  title        = {Pith review of: SuperMapNet for Long-Range and High-Accuracy Vectorized HD Map Construction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5JPMCEEF}},
  note         = {Machine review of arXiv:2505.13856}
}
read the original abstract

Vectorized HD map is essential for autonomous driving. Remarkable work has been achieved in recent years, but there are still major issues: (1) in the generation of the BEV features, single modality-based methods are of limited perception capability, while direct concatenation-based multi-modal methods fail to capture synergies and disparities between different modalities, resulting in limited ranges with feature holes; (2) in the classification and localization of map elements, only point information is used without the consideration of element infor-mation and neglects the interaction between point information and element information, leading to erroneous shapes and element entanglement with low accuracy. To address above issues, we introduce SuperMapNet for long-range and high-accuracy vectorized HD map construction. It uses both camera images and LiDAR point clouds as input, and first tightly couple semantic information from camera images and geometric information from LiDAR point clouds by a cross-attention based synergy enhancement module and a flow-based disparity alignment module for long-range BEV feature generation. And then, local features from point queries and global features from element queries are tightly coupled by three-level interactions for high-accuracy classification and localization, where Point2Point interaction learns local geometric information between points of the same element and of each point, Element2Element interaction learns relation constraints between different elements and semantic information of each elements, and Point2Element interaction learns complement element information for its constituent points. Experiments on the nuScenes and Argoverse2 datasets demonstrate superior performances, surpassing SOTAs over 14.9/8.8 mAP and 18.5/3.1 mAP under hard/easy settings, respectively. The code is made publicly available1.

Figures

Figures reproduced from arXiv: 2505.13856 by the authors.

Figure 1
Figure 1. Visualization comparison between rasterized and vectorized HD map. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The overall architecture of SuperMapNet. Semantic information from camera images and geometric information from LiDAR point clouds are tightly [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. The evolution of deep learning-based HD map construction methods with multi-modalities. According to di [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Semantic and geometric coupling (SGC) module. The synergy enhancement is aimed to mining the relations and complementarity between camera BEV [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Point and element coupling (PEC) by interactions at three levels, where Point2Point interaction for local information learning between points of the same [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Global view of nuScenes dataset. Dm l+1 = Dm l + so f tmax Qm l (Km,i l ) T √ dk ! Dm,i l (20) where Dm,i l is the i-th point’s descriptor of the element m in l-th layer, Dm l is the descriptor of the element m in l-th layer. Q and K are obtained by two fully-connected…
Figure 7
Figure 7. Figure 7: Challenges on Argoverse2 dataset. 5. Ablation Study 5.1. Accuracy of different element types The significant differences in the shapes of different types of elements bring challenges for modeling map elements in a uni￾fied representation. Tab. 3 lists how the proposed …
Figure 8
Figure 8. Figure 8: Visualization comparison between different modules of SuperMapNet on nuScenes dataset, where erroneous shapes are circled in blue and entanglement between elements are in yellow. Road boundaries are colored in green, while lane dividers and pedestrian crossings are in …
Figure 9
Figure 9. Figure 9: Visualization of SuperMapNet on Argoverse2 dataset. Road boundaries are colored in green, while lane dividers and pedestrian crossings are in red and [PITH_FULL_IMAGE:figures/full_fig_p013_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

23 extracted references · 16 canonical work pages

  1. [5]

    15922–15928

    Mb- fusion: A new multi-modal bev feature fusion method for hd map construc- tion, in: 2024 IEEE International Conference on Robotics and Automation (ICRA), pp. 15922–15928. Hu, H., Wang, F., Wang, Y ., Hu, L., Xu, J., Zhang, Z.,

    work page 2024

  2. [6]

    Admap: Anti- disturbance framework for vectorized hd map construction, in: Computer Vision – ECCV 2024, p. 311–326. Jia, P., Wen, T., Luo, Z., Yang, M., Jiang, K., Liu, Z., Tang, X., Lei, Z., Cui, L., Zhang, B., Sheng, K., Yang, D.,

    work page 2024

  3. [10]

    9992–10002

    Swin transformer: Hierarchical vision transformer using shifted windows, in: 2021 IEEE/CVF International Conference on Computer Vision (ICCV), pp. 9992–10002. Liu, Z., Tang, H., Amini, A., Yang, X., Mao, H., Rus, D.L., Han, S., 2023b. Bevfusion: Multi-task multi-sensor fusion with unified bird’s-eye view rep- resentation, in: 2023 IEEE International Confe...

    work page 2021

  4. [11]

    Leveraging enhanced queries of point sets for vectorized map construction, in: Computer Vision – ECCV 2024: 18th European Conference, Milan, Italy, September 29–Oc- tober 4, 2024, Proceedings, Part LVII, p. 461–477. Luo, K.Z., Weng, X., Wang, Y ., Wu, S., Li, J., Weinberger, K.Q., Wang, Y ., Pavone, M.,

    work page 2024

  5. [12]

    4029–4035

    Augmenting lane perception and topology understand- ing with standard definition navigation maps, in: 2024 IEEE International Conference on Robotics and Automation (ICRA), pp. 4029–4035. Luo, Z., Gao, L., Xiang, H., Li, J.,

    work page 2024

  6. [14]

    Qiao, L., Ding, W., Qiu, X., Zhang, C., 2023a

    Lift, splat, shoot: Encoding images from arbi- trary camera rigs by implicitly unprojecting to 3d, in: Proceedings of the European Conference on Computer Vision. Qiao, L., Ding, W., Qiu, X., Zhang, C., 2023a. End-to-end vectorized hd-map construction with piecewise b ´ezier curve, in: 2023 IEEE /CVF Conference on Computer Vision and Pattern Recognition (C...

    arXiv 2023

  7. [15]

    CoRR abs/2311.10517

    Mind the map! accounting for existing map information when estimating online hdmaps from sensor. CoRR abs/2311.10517. arXiv:2306.10301. Wang, S., Jia, F., Mao, W., Liu, Y ., Zhao, Y ., Chen, Z., Wang, T., Zhang, C., Zhang, X., Zhao, F.,

    arXiv

  8. [17]

    Xiong, X., Liu, Y ., Yuan, T., Wang, Y ., Wang, Y ., Zhao, H.,

    Argoverse 2: Next generation datasets for self-driving percep- tion and forecasting, in: Proceedings of the Neural Information Processing Systems Track on Datasets and Benchmarks (NeurIPS Datasets and Bench- marks 2021). Xiong, X., Liu, Y ., Yuan, T., Wang, Y ., Wang, Y ., Zhao, H.,

    work page 2021

Show all 23 references
  1. [18]

    17535–17544

    Neural map prior for autonomous driving, in: 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 17535–17544. doi: 10.1109/ CVPR52729.2023.01682. Xu, Z., K. Wong, K.Y ., Zhao, H.,

    2023

  2. [19]

    Insmapper: Exploring inner-instance information for vectorized hd mapping, in: Computer Vision – ECCV 2024, p. 296–312. Yan, Y ., Mao, Y ., Li, B.,

    2024

  3. [20]

    Widthformer: To- ward efficient transformer-based bev view transformation

    Yang, C., Lin, T., Huang, L., Crowley, E.J., 2024a. Widthformer: To- ward efficient transformer-based bev view transformation. arXiv preprint arXiv:2401.03836 . Yang, J., Jiang, M., Yang, S., Tan, X., Li, Y ., Ding, E., Wang, H., Wang, J., 2024b. Mgmapnet: Multi-granularity re...

    arXiv

  4. [21]

    ISPRS Journal of Photogramme- try and Remote Sensing 204, 27–41

    Generating evidential bev maps in continuous driving space. ISPRS Journal of Photogramme- try and Remote Sensing 204, 27–41. doi:https://doi.org/10.1016/j. isprsjprs.2023.08.013. Zhang, G., Lin, J., Wu, S., Song, Y ., Luo, Z., Xue, Y ., Lu, S., Wang, Z., 2024a. Online map vect...

    2023 doi

  5. [22]

    Online vectorized hd map construction using geometry, in: Computer Vision – ECCV 2024, pp. 73–90. Zhou, Y ., Tuzel, O.,

    2024

  6. [24]

    15396–15406

    Himap: Hybrid representation learning for end-to-end vectorized hd map construction, in: 2024 IEEE/CVF Conference on Computer Vision and Pat- tern Recognition (CVPR), pp. 15396–15406. Zhu, X., Li, W., Wang, Z., Li, M., Wang, X., Dai, J.,

    2024

  7. [25]

    Deformable detr: Deformable transformers for end-to-end object detection, in: 9th Interna- tional Conference on Learning Representations, ICLR 2021, Virtual Event, Austria, May 3-7,

    2021

  8. [2018]

    4490–4499

    V oxelnet: End-to-end learning for point cloud based 3d object detection, in: 2018 IEEE /CVF Conference on Computer Vision and Pattern Recognition, pp. 4490–4499. Zhou, Y ., Zhang, H., Yu, J., Yang, Y ., Jung, S., Park, S.I., Yoo, B.,

    2018

  9. [2019]

    12689–12697

    Pointpillars: Fast encoders for object detection from point clouds, in: 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 12689–12697. Li, Q., Wang, Y ., Wang, Y ., Zhao, H.,

    2019

  10. [2020]

    11618–11628

    nuscenes: A multimodal dataset for autonomous driving, in: 2020 IEEE /CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 11618–11628. 11 Figure 8: Visualization comparison between different modules of SuperMapNet on nuScenes dataset, where erroneous shapes a...

    2020

  11. [2021]

    9630–9640

    Emerging properties in self-supervised vision transformers, in: 2021 IEEE/CVF International Conference on Computer Vision (ICCV), pp. 9630–9640. doi: 10.1109/ICCV48922.2021.00951. Chitta, K., Prakash, A., Jaeger, B., Yu, Z., Renz, K., Geiger, A.,

    2021

  12. [2022]

    4628–4634

    Hdmapnet: An online hd map construction and evaluation framework, in: 2022 International Con- ference on Robotics and Automation (ICRA), p. 4628–4634. doi:10.1109/ ICRA46639.2022.9812383. Li, S., Li, G., Wang, L., Qin, Y .,

    2022

  13. [2023]

    3649–3659

    Pivotnet: Vectorized pivot learn- ing for end-to-end hd map construction, in: 2023 IEEE /CVF International Conference on Computer Vision (ICCV), pp. 3649–3659. Dong, H., Gu, W., Zhang, X., Xu, J., Ai, R., Lu, H., Kannala, J., Chen, X.,

    2023

  14. [2024]

    9056–9062

    Superfusion: Multilevel lidar-camera fusion for long-range hd map genera- tion, in: 2024 IEEE International Conference on Robotics and Automation (ICRA), pp. 9056–9062. Hao, X., Zhang, H., Yang, Y ., Zhou, Y ., Jung, S., Park, S.I., Yoo, B.,

    2024

  15. [2025]

    (Eds.), Computer Vision – ECCV 2024, Springer Nature Switzerland, Cham

    Stream query denoising for vectorized hd-map construction, in: Leonardis, A., Ricci, E., Roth, S., Russakovsky, O., Sat- tler, T., Varol, G. (Eds.), Computer Vision – ECCV 2024, Springer Nature Switzerland, Cham. pp. 203–220. Wilson, B., Qi, W., Agarwal, T., Lambert, J., Singh...

    2024

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.