HD map Instance Matching Dataset

High-Definition Lane-Instance Matching Dataset This dataset provides lane-level map “instances” (center-lines and lane-dividers) for four urban test areas—Boston Seaport, Singapore Holland Village, Singapore One North, and Singapore Queenstown—together with simulated per-frame local inference outputs and ground-truth correspondence data. It is designed to support research on incremental HD-map update, instance-level matching, and topological/ geometric consistency evaluation. Contents: Raw OSM maps (.osm) for each area (Derived from the work: Lanelet2 for nuScenes) Ground-truth instances extracted from OSM: Center-line and lane-divider point sequences Inter-instance connectivity graphs (edge lists) Left/right cross-adjacency indices Serialized Python pickles (.pkl) Simulated local inference frames for each map: Per-frame “perception box” polygons Noise-augmented partial segments sampled from the full map Use Cases: Benchmarking instance-level matching algorithms Evaluating geometric/topological consistency (Average Precision, F1) Developing robust incremental mapping pipelines (fusion, deletion, extension) Visualizing HD-map update over time License: MIT