AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Failure Modes for Deep Learning-Based Online Mapping: How to Measure and Address Them

Michael Hubbertz, Qi Han, Tobias Meisen

CVPR 2026 · Label-efficient and robust learning › Recognition and long-tail · approach: Dense features and foundation priors · 0 citations (OpenAlex)

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 changedWhereMeasuredEffect
Use geographically disjoint split instead of originalMapTRv2 · nuScenesmAP: 60.95 -> 24.96 (geo. [19]), 28.53 (geo. [37])Hurt
Use geographically disjoint split instead of originalMapTRv2 · Argoverse 2 (2 Hz)mAP: 63.97 -> 49.53 (geo. [19]), 57.61 (geo. [37])Hurt
Use geometry-maximizing split instead of originalMapTRv2 · Argoverse 2 (2 Hz)mAP: 63.97 -> 34.13Hurt
Higher geometric diversity of training set (geo. [37] vs [19])MapTRv2 · nuScenesmAP: 24.96 -> 28.53; geomdiv(T) 80.6 km -> 90.2 kmImproved
MST-based sparsification with low threshold (0.1-1) (ablation)MapTRv2 · nuScenes/Argoverse 2 validationmAPImproved
MST-based sparsification with high threshold (2-10) (ablation)MapTRv2 · nuScenes/Argoverse 2 validationmAPHurt
Random subsampling instead of MST sparsification (ablation)MapTRv2 · nuScenes/Argoverse 2 originalmAPHurt
Architecture choice MapQR vs MapTRv2MapQR · nuScenes originalOloc (%): 24.73 -> 57.07Hurt
Fréchet-based median M instead of mAP on small subsetsevaluation measure · nuScenesmetric validityImproved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy