Reproduce “TopoSD: Topology-Enhanced Lane Segment Perception with SDMap Prior” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Sen Yang, Minyue Jiang, Ziwei Fan, Xiaolu Xie, Xiao Tan, Yingying Li, Errui Ding, Liang Wang and 1 more
Baidu Inc. researchers developed TOPOSD, a perception system that integrates Standard Definition Maps (SDMaps) as prior information to improve lane segment geometry and topology prediction. The system achieved a new state-of-the-art performance on the OpenLane-V2 dataset, reaching 40.2% mAP and 34.5% TOP_lsls.
Key insights
- Integrating SDMap priors at an early stage of processing (pre-fusion into BEV queries and features) provides crucial contextual information, acting as an effective 'prompt' to guide perception.
- The mutual relationship between lane geometry and topology can be effectively exploited through an iterative decoding process, leading to more consistent and accurate predictions for both aspects.
- Training with noisy SDMaps significantly enhances model robustness to real-world inaccuracies in SDMaps, demonstrating a balanced reliance on both visual features and map priors.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.