Research map / Label-efficient and robust learning / Semi-supervised learning
PseudoMapTrainer: Learning Online Mapping without HD Maps
Christian Löwens, Thorben Funke, Jingchao Xie, Alexandru Paul Condurache
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 |
|---|---|---|---|
| lower coverage threshold from 50% to 30% without mask-aware training (ablation) | MapTRv2 · Ablation (single-trip pseudo-labels) | mean AP: 6.7 -> 5.3 | Hurt |
| add rendering and direction losses (ablation) | MapTRv2 (tau_m 30%) · Ablation (single-trip pseudo-labels) | mean AP: 5.3 -> 6.1 | Mixed |
| add PV segmentation loss without projection (ablation) | MapTRv2 · Ablation (single-trip pseudo-labels) | mean AP: 6.1 -> 6.4 | Improved |
| add mask-aware assignment and loss (ablation) | MapTRv2 · Ablation (single-trip pseudo-labels) | mean AP: 6.4 -> 8.2 | Improved |
| add mask-aware assignment and loss (ablation) | MapTRv2 · Ablation (single-trip pseudo-labels) | road boundary AP: 4.7 -> 8.3 | Improved |
| train with multi-trip instead of single-trip pseudo-labels | MapTRv2 + PseudoMapTrainer · nuScenes val | mean AP: 8.2 -> 13.2 | Improved |
| add PseudoMapTrainer to naive pseudo-label training | MapTRv2 (multi-trip) · nuScenes val | mean AP: 10.5 -> 13.2 | Improved |
| use multi-trip pseudo-labels instead of single-trip | pseudo-label generation · Pseudo-label BEV coverage | coverage ratio: 40.0% -> 65.5% | Improved |
| increase BEV resolution from 5 to 40 pixel/m (ablation) | pseudo-label pipeline · Pseudo-label quality, observed area | lane divider AP: 3.1 -> 6.5 | Improved |
| increase dilation kernel size from 1 to 25 (ablation) | pseudo-label pipeline · Pseudo-label quality, observed area | lane divider AP: 3.7 -> 6.7 | Improved |
| use very small distance step epsilon 1 cm instead of 5 cm (ablation) | pseudo-label pipeline · Pseudo-label quality, observed area | pedestrian crossing AP: 23.6 -> 20.8 | Hurt |
| increase BEV resolution from 5 to 40 pixel/m (ablation) | pseudo-label pipeline · Pseudo-label quality, observed area | pedestrian crossing AP: 24.0 -> 22.8 | Hurt |
Closest papers by meaning
- Accelerating Online Mapping and Behavior Prediction via Direct BEV Feature Attention (ECCV 2024 · similarity 0.87)
- Patched Line Segment Learning for Vector Road Mapping (AAAI 2024 · similarity 0.86)
- Unveiling the Hidden: Online Vectorized HD Map Construction with Clip-Level Token Interaction and Propagation (NeurIPS 2024 · similarity 0.85)
- MapExpert: Online HD Map Construction with Simple and Efficient Sparse Map Element Expert (AAAI 2025 · similarity 0.85)
- SDTagNet: Leveraging Text-Annotated Navigation Maps for Online HD Map Construction (NeurIPS 2025 · similarity 0.85)
- Swift-Mapping: Online Neural Implicit Dense Mapping in Urban Scenes (AAAI 2024 · similarity 0.84)
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.
