Research map / Architectures and efficiency / RNNs and normalization
MapTR: Structured Modeling and Learning for Online Vectorized HD Map Construction
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Replace fixed point ordering with permutation-equivalent point-set modeling (ablation) | MapTR-tiny · nuScenes val | mAP: 44.4 -> 50.3; +5.9 mAP | Helped |
| Add edge direction loss with weight beta 5e^{-3} rather than removing it (ablation) | MapTR-tiny · nuScenes val | mAP: 48.2 -> 50.3 | Helped |
| Increase edge direction loss weight to 1e^{-2} instead of the default 5e^{-3} (ablation) | MapTR-tiny · nuScenes val | mAP: 50.3 -> 48.3 | Hurt |
| Use GKT rather than the tested alternative 2D-to-BEV transformations (ablation) | MapTR-tiny · nuScenes val | mAP: 46.2 (IPM), 49.5 (LSS), 49.7 (Deform. Atten.), 50.3 (GKT) | Helped |
| Use 20 points rather than 10 or 40 points per map element (ablation) | MapTR-tiny · nuScenes val | mAP: 48.0 (10 points), 50.3 (20 points), 50.0 (40 points) | Mixed |
| Increase map element queries from 25 to 50 (ablation) | MapTR-tiny · nuScenes val | mAP: 41.6 -> 50.3 | Helped |
| Increase map element queries from 50 to 75 (ablation) | MapTR-tiny · nuScenes val | mAP: 50.3 -> 52.2 | Helped |
| Increase decoder layers from 1 to 6 (ablation) | MapTR-tiny · nuScenes val | mAP: 29.1 -> 50.3 | Helped |
| Increase decoder layers from 6 to 8 (ablation) | MapTR-tiny · nuScenes val | mAP: 50.3 -> 47.6 | Hurt |
| Use point2point rather than Chamfer distance for instance-level position matching (ablation) | MapTR-tiny · nuScenes val | mAP: 47.5 -> 50.3 | Helped |
| Add LiDAR input to camera input (ablation) | MapTR-tiny · nuScenes val | mAP: 50.3 -> 62.5 | Helped |
| Add camera translation deviation with standard deviation 0.1m (ablation) | MapTR-tiny · nuScenes val | mAP: 50.3 -> 49.0 | Hurt |
Closest papers by meaning
- Unveiling the Hidden: Online Vectorized HD Map Construction with Clip-Level Token Interaction and Propagation (NeurIPS 2024 · similarity 0.89)
- MapExpert: Online HD Map Construction with Simple and Efficient Sparse Map Element Expert (AAAI 2025 · similarity 0.89)
- Uncertainty-Instructed Structure Injection for Generalizable HD Map Construction (CVPR 2025 · similarity 0.87)
- Compact HD Map Construction via Douglas-Peucker Point Transformer (AAAI 2024 · similarity 0.87)
- HIMap: HybrId Representation Learning for End-to-end Vectorized HD Map Construction (CVPR 2024 · similarity 0.87)
- InteractionMap: Improving Online Vectorized HDMap Construction with Interaction (CVPR 2025 · similarity 0.87)
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.