Reproduce “IC-Mapper: Instance-Centric Spatio-Temporal Modeling for Online Vectorized Map Construction” 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
Jiangtong Zhu, Zhao Yang, Yinan Shi, Jianwu Fang, Jianru Xue
IC-Mapper presents an end-to-end instance-centric framework for online vectorized global map construction from continuous visual data, addressing the limited spatial scalability of existing methods. It integrates detection, temporal association, and spatial fusion to enable real-time global map updates, demonstrating up to 1.6% mAP improvement over StreamMapNet in detection and superior mapping performance compared to other fusion techniques.
Key insights
- An instance-centric approach, combining detection, tracking, and fusion, can overcome the spatial limitations of fixed-range BEV map construction methods and enable real-time global map updates.
- Temporal association using a tracking memory buffer and similarity-based matching (geometric and feature) is essential for maintaining consistent map instance identities across frames.
- Spatial fusion, incorporating class-specific merging strategies like Bezier curve fitting for linear elements and polygon union for area elements, effectively integrates new observations into a smooth, spatially consistent global map.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.