Seg-LaneDet: 3D Lane Detection from Monocular Images with 2D Segmentation

Yu Wang, Yifan Jiao, Tian He, Zhenming Zhang, weijie qiu, Zhuqing Jiang

Keywords: 3D Lane Detection, Autonomous Driving, Computer Vision

TL;DR: We propose a simple yet effective 3D lane detector with the front-view segmentation

Abstract: Monocular 3D lane detection is a fundamental yet challenging task in autonomous driving. Recent advancements primarily rely on constructing 3D surrogates from monocular images and camera parameters. However, misalignment is introduced in current methods due to the lack of dense depth information in datasets, coupled with the inherent depth ambiguity of monocular images. To address this issue, we propose Seg-LaneDet, a simple but effective end-to-end 3D lane detector. We frame the task of 3D lane detection as an elevation from 2D to 3D detection. Specifically, we leverage a pre-trained 2D lane detector to obtain instance segmentation of lanes, of which the segmentation maps serve as the sole prior for the 2D-to-3D module. This allows us to achieve a straightforward 3D lane representation based on front-view segmentation maps. Our method demonstrates comparable performance to state-of-the-art (SOTA) F1 scores on the OpenLane and the Apollo datasets.

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: 9870