Research map / Architectures and efficiency / RNNs and normalization
MapTRv2: An End-to-End Framework for Online Vectorized HD Map Construction
Bencheng Liao, Shaoyu Chen, Yunchi Zhang, Bo Jiang, Qian Zhang, Wenyu Liu, Chang Huang, Xinggang Wang
Read the paperCheck your idea against 137,217 papers
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 permutation-equivalent shape modeling (ablation) | MapTRv2-R50 · nuScenes val (24 epochs) | mAP: 57.4 -> 61.5 | Improved |
| add decoupled self-attention instead of vanilla (ablation) | MapTRv2-R50 · nuScenes val (24 epochs) | mAP: 57.1 -> 57.6 | Improved |
| use only inter-instance self-attention instead of vanilla (ablation) | MapTRv2-R50 · nuScenes val (24 epochs) | mAP: 57.1 -> 56.8 | No clear effect |
| replace BEV cross-attention with PV cross-attention (ablation) | MapTRv2-R50 · nuScenes val (24 epochs) | mAP: 61.5 -> 49.5 | Hurt |
| stack PV cross-attention on BEV (hybrid) (ablation) | MapTRv2-R50 · nuScenes val (24 epochs) | mAP: 61.5 -> 60.5 | Hurt |
| stack PV cross-attention on BEV (hybrid) (ablation) | MapTRv2-R50 · Argoverse2 val (3D map) | mAP: 64.7 -> 65.6 | Improved |
| add depth, PV seg and BEV seg dense losses (ablation) | MapTRv2-R50 · nuScenes val (24 epochs) | mAP: 56.6 -> 61.5 | Improved |
| add auxiliary one-to-many matching with K=6 (ablation) | MapTRv2-R50 · nuScenes val (24 epochs) | mAP: 57.6 -> 61.5 | Improved |
| increase number of points from 20 to 40 (ablation) | MapTRv2-R50 · nuScenes val (24 epochs) | mAP: 61.5 -> 56.9 | Hurt |
| increase decoder layers from 1 to 2 (ablation) | MapTRv2-R50 · nuScenes val (24 epochs) | mAP: 39.5 -> 52.6 | Improved |
| increase decoder layers from 6 to 8 (ablation) | MapTRv2-R50 · nuScenes val (24 epochs) | mAP: 61.5 -> 61.3 | No clear effect |
| replace GKT with LSS PV-to-BEV transformation | MapTR-R50 · nuScenes val (24 epochs) | mAP: 50.3 -> 50.1 | No clear effect |
Closest papers by meaning
- MapExpert: Online HD Map Construction with Simple and Efficient Sparse Map Element Expert (AAAI 2025 · similarity 0.91)
- Unveiling the Hidden: Online Vectorized HD Map Construction with Clip-Level Token Interaction and Propagation (NeurIPS 2024 · similarity 0.90)
- InteractionMap: Improving Online Vectorized HDMap Construction with Interaction (CVPR 2025 · similarity 0.89)
- MGMap: Mask-Guided Learning for Online Vectorized HD Map Construction (CVPR 2024 · similarity 0.88)
- HIMap: HybrId Representation Learning for End-to-end Vectorized HD Map Construction (CVPR 2024 · similarity 0.88)
- Compact HD Map Construction via Douglas-Peucker Point Transformer (AAAI 2024 · 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.
