TY - GEN
T1 - Online Vectorized HD Map Construction Using Geometry
AU - Zhang, Zhixin
AU - Zhang, Yiyuan
AU - Ding, Xiaohan
AU - Jin, Fusheng
AU - Yue, Xiangyu
N1 - Publisher Copyright:
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.
PY - 2025
Y1 - 2025
N2 - Online vectorized High-Definition (HD) map construction is critical for downstream prediction and planning. Recent efforts have built strong baselines for this task, however, geometric shapes and relations of instances in road systems are still under-explored, such as parallelism, perpendicular, rectangle-shape, etc. In our work, we propose GeMap (Geometry Map), which end-to-end learns Euclidean shapes and relations of map instances beyond fundamental perception. Specifically, we design a geometric loss based on angle and magnitude clues, robust to rigid transformations of driving scenarios. To address the limitations of the vanilla attention mechanism in learning geometry, we propose to decouple self-attention to handle Euclidean shapes and relations independently. GeMap achieves new state-of-the-art performance on the nuScenes and Argoverse 2 datasets. Remarkably, it reaches a 71.8% mAP on the large-scale Argoverse 2 dataset, outperforming MapTRv2 by +4.4% and surpassing the 70% mAP threshold for the first time. Code is available at https://github.com/cnzzx/GeMap.
AB - Online vectorized High-Definition (HD) map construction is critical for downstream prediction and planning. Recent efforts have built strong baselines for this task, however, geometric shapes and relations of instances in road systems are still under-explored, such as parallelism, perpendicular, rectangle-shape, etc. In our work, we propose GeMap (Geometry Map), which end-to-end learns Euclidean shapes and relations of map instances beyond fundamental perception. Specifically, we design a geometric loss based on angle and magnitude clues, robust to rigid transformations of driving scenarios. To address the limitations of the vanilla attention mechanism in learning geometry, we propose to decouple self-attention to handle Euclidean shapes and relations independently. GeMap achieves new state-of-the-art performance on the nuScenes and Argoverse 2 datasets. Remarkably, it reaches a 71.8% mAP on the large-scale Argoverse 2 dataset, outperforming MapTRv2 by +4.4% and surpassing the 70% mAP threshold for the first time. Code is available at https://github.com/cnzzx/GeMap.
KW - Geometry Representation
KW - Geometry-Decoupled Attention
KW - HD Map Construction
UR - http://www.scopus.com/inward/record.url?scp=85209377745&partnerID=8YFLogxK
U2 - 10.1007/978-3-031-72967-6_5
DO - 10.1007/978-3-031-72967-6_5
M3 - Conference contribution
AN - SCOPUS:85209377745
SN - 9783031729669
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 73
EP - 90
BT - Computer Vision – ECCV 2024 - 18th European Conference, Proceedings
A2 - Leonardis, Aleš
A2 - Ricci, Elisa
A2 - Roth, Stefan
A2 - Russakovsky, Olga
A2 - Sattler, Torsten
A2 - Varol, Gül
PB - Springer Science and Business Media Deutschland GmbH
T2 - 18th European Conference on Computer Vision, ECCV 2024
Y2 - 29 September 2024 through 4 October 2024
ER -