UMVMap: Improving Vectorized Map Construction via Multi-vehicle Perspectives
Keywords: multi-vehicle, autonomous driving, vectorized map construction
TL;DR: UMVMap: Improving Vectorized Map Construction via Multi-vehicle Perspectives
Abstract: Prevalent vectorized map construction pipelines predominantly follow an end-to-end DETR-based paradigm. While these methods have achieved significant advancements, they are limited by their reliance on data from a single ego vehicle, which restricts their effectiveness and can lead to perceptual uncertainty in handling complex environmental scenarios. To address this limitation, we introduce a novel framework: Uncertainty-aware Multi-Vehicle Vectorized Map Construction (UMVMap). This framework effectively mitigates uncertainties by leveraging relevant non-ego information. UMVMap comprises two essential components: the Uncertainty-aware Multi-Vehicle Vectorized Map Construction Network (UMVMap-Net), which optimally integrates data from multiple vehicles, and the Uncertainty-aware Non-ego Vehicle Selection (UNVS) strategy, which identifies and incorporates the most informative non-ego data to minimize uncertainty. Comprehensive evaluations on the nuScenes dataset demonstrate that UMVMap significantly outperforms the single-vehicle MapTRv2 baseline by a margin of 9.1\% and 9.9\% respectively on the full and partial validation sets, with each of its components proving to be both effective and robust.
Primary Area: applications to computer vision, audio, language, and other modalities
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 331
Loading