Research map / Label-efficient and robust learning / Knowledge distillation
AMap: Distilling Future Priors for Ahead-Aware Online HD Map Construction
Ruikai Li, Xinrun Li, Mengwei Xie, Hao Shan, Shoumeng Qiu, Xinyuan Chang, Yizhe Fan, Feng Xiong, Han Jiang, Yilong Ren, Haiyang Yu, Mu Xu and 3 more
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 |
|---|---|---|---|
| mask forward map region fully before trajectory prediction (ablation) | MapTR maps with HiVT predictor · downstream trajectory prediction | minFDE: 0.7919 -> 0.9024 (+13.95%) | Hurt |
| mask backward map region before trajectory prediction (ablation) | MapTR maps with HiVT predictor · downstream trajectory prediction | Miss Rate: 0.0854 -> 0.0807 at 70% mask | No clear effect |
| distill from a future-frame teacher into current-frame student (AMap) | MapTracker R18 student · nuScenes val | A-mAP: 64.63 -> 66.28 | Improved |
| add AMap distillation to MapTR R18 | MapTR R18 · nuScenes val | A-mAP: 36.34 -> 51.29 | Improved |
| add AMap distillation on Argoverse 2 | MapTracker R18 student · Argoverse 2 val | A-mAP: 65.29 -> 67.81 | Improved |
| use BEVDistill or MapDistill instead of AMap for future-to-static distillation | MapTracker R18 student · nuScenes val | mAP: 62.81 -> 57.97 (BEVDistill), 57.82 (MapDistill) | Hurt |
| distill both basic and refined BEV features (ablation) | AMap student · nuScenes val | mAP: 62.81 -> 63.47 | Improved |
| distill only basic BEV features (ablation) | AMap student · nuScenes val | mAP: 62.81 -> 62.66 | No clear effect |
| add query-level to BEV-level distillation (ablation) | AMap student · nuScenes val | mAP: 63.47 -> 64.49 | Improved |
| refined-BEV-only distillation (ablation) | AMap student · nuScenes val | mAP vs A-mAP/R-mAP: mAP 62.81 -> 62.72, A-mAP 64.63 -> 64.68, R-mAP 64.04 -> 64.78 | Mixed |
| use Hungarian matching on query embeddings (MSE or cosine) instead of logits KL (ablation) | AMap query distillation · nuScenes val | mAP: 63.63 -> 62.99 (MSE), 62.43 (cosine) | Hurt |
| use MapDistill dummy or Top-K query distillation (ablation) | AMap query distillation · nuScenes val | mAP: 63.63 -> 57.63 (dummy), 56.61 (Top K) | Hurt |
Closest papers by meaning
- Future-Aware End-to-End Driving: Bidirectional Modeling of Trajectory Planning and Scene Evolution (NeurIPS 2025 · similarity 0.88)
- MapExpert: Online HD Map Construction with Simple and Efficient Sparse Map Element Expert (AAAI 2025 · similarity 0.88)
- FutureSightDrive: Thinking Visually with Spatio-Temporal CoT for Autonomous Driving (NeurIPS 2025 · similarity 0.88)
- Leveraging SD Map to Augment HD Map-based Trajectory Prediction (CVPR 2025 · similarity 0.87)
- SDTagNet: Leveraging Text-Annotated Navigation Maps for Online HD Map Construction (NeurIPS 2025 · similarity 0.87)
- Producing and Leveraging Online Map Uncertainty in Trajectory Prediction (CVPR 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.
