AI Paper Atlas

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

Stability Under Scrutiny: Benchmarking Representation Paradigms for Online HD Mapping

Hao Shan, Ruikai Li, Han Jiang, Yizhe Fan, Ziyang Yan, Bohan Li, Xiaoshuai Hao, Hao Zhao, Zhiyong Cui, Yilong Ren, Haiyang Yu

ICLR 2026 · Label-efficient and robust learning › Recognition and long-tail · approach: Dense features and foundation priors · 8 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
add LiDAR fusion (ablation)MapTR · nuScenes valmAS: 71.6 -> 74.0Improved
add LiDAR fusion (ablation)GeMap · nuScenes valmAS: 74.7 -> 71.8Hurt
add LiDAR fusion (ablation)MapTR and GeMap · nuScenes valmAP: 44.1 -> 62.8; 62.7 -> 66.5Improved
swap BEV encoder (BEVFormer, GKT, BEVPool) (ablation)MapTR · nuScenes valmAS: 71.3 to 71.9 (mAP 41.6 to 50.1)No clear effect
add temporal fusion (GKT encoder) (ablation)MapTR · nuScenes valmAS: 71.6 -> 66.6Hurt
add temporal fusion (BEVFormer encoder) (ablation)MapTR · nuScenes valmAS: 71.3 -> 73.0Improved
add native temporal modeling (ablation)StreamMapNet · nuScenes valmAS: 83.8 -> 91.9Improved
add temporal fusion (ablation)MapTracker (R50) · nuScenes valmAS: mAS 90.8 -> 90.4; mAP 68.3 -> 75.95Mixed
add training map priors (ablation)HRMapNet · nuScenes valmAS: mAP 67.2 -> 83.6; mAS 75.9 -> 76.7Mixed
use stronger 2D backbone (R18 to R50) (ablation)MapTR · nuScenes valmAS: 72.8 -> 71.6Hurt
use stronger 2D backbone (R18 to R50) (ablation)MapQR · nuScenes valmAS: 74.1 -> 77.8Improved
train longer (24 to 110 epochs) (ablation)MapTR-50 · nuScenes valmAS: 71.6 -> 68.2Hurt

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