Research map / Label-efficient and robust learning / Recognition and long-tail
Enhancing Vectorized Map Perception with Historical Rasterized Maps
Xiaoyu Zhang, Guangwei Liu, Zihao Liu, Ningyi Xu, Yunhui Liu, Ji Zhao
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add historical rasterized map (HRMapNet) to StreamMapNet | StreamMapNet · nuScenes val (24 epochs) | mAP: 60.4 -> 66.3 | Improved |
| add historical rasterized map (HRMapNet) to MapTRv2 with 110 epochs | MapTRv2 · nuScenes val (110 epochs) | mAP: 68.7 -> 73.6 | Improved |
| add historical rasterized map | StreamMapNet and MapTRv2 · Argoverse 2 val | mAP: 61.5 -> 64.3 (StreamMapNet); 64.3 -> 68.3 (MapTRv2) | Improved |
| add historical map without temporal fusion modules on location-disjoint splits | StreamMapNet · nuScenes and Argoverse 2 new splits | mAP: 33.9 -> 37.1 (nuScenes); 58.1 -> 61.5 (Argoverse 2) | Improved |
| add BEV feature aggregation with local map (ablation) | MapTRv2 · nuScenes val (24 epochs) | mAP: 61.5 -> 64.6 | Improved |
| add query initialization with map prior embeddings (ablation) | MapTRv2 + feature aggregation · nuScenes val (24 epochs) | mAP: 64.6 -> 67.2 | Improved |
| downsample local map to 0.6 m before extracting prior embeddings (ablation) | query initialization · nuScenes val | mAP / GPU memory: mAP 66.3 -> 67.2; memory 64.94 GB -> 31.41 GB | Improved |
| add Gaussian ego-pose noise at test time (ablation) | HRMapNet · nuScenes val | mAP: 67.2 -> 66.9 (0.1 m); 63.8 worst case | No clear effect |
| start from validation-built or training-built global map instead of empty (ablation) | HRMapNet · nuScenes val | mAP: 67.2 -> 72.6 (validation map); 67.2 -> 83.7 (training map) | Improved |
| use query initialization with fewer decoder layers (ablation) | HRMapNet · nuScenes val (24 epochs) | mAP: 64.6 (6 layers, no query init) -> 65.4 (4 layers, query init) | Improved |
| add historical map under adverse conditions (night, rain) | MapTRv2 · nuScenes night and rainy | mAP: night 39.6 -> 42.6; rainy 50.8 -> 62.6 | Improved |
| HRMapNet inference cost | MapTRv2 · nuScenes val | FPS: 19.6 -> 17.0 | Hurt |
Closest papers by meaning
- PriorDrive: Enhancing Online HD Mapping with Unified Vector Priors (AAAI 2026 · similarity 0.90)
- MapExpert: Online HD Map Construction with Simple and Efficient Sparse Map Element Expert (AAAI 2025 · similarity 0.90)
- Map It Anywhere: Empowering BEV Map Prediction using Large-scale Public Datasets (NeurIPS 2024 · similarity 0.87)
- SDTagNet: Leveraging Text-Annotated Navigation Maps for Online HD Map Construction (NeurIPS 2025 · similarity 0.87)
- InteractionMap: Improving Online Vectorized HDMap Construction with Interaction (CVPR 2025 · similarity 0.86)
- Uncertainty-Instructed Structure Injection for Generalizable HD Map Construction (CVPR 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
