Home

Awesome

ADMap: Anti-disturbance framework for reconstructing online vectorized HD map

Due to confidentiality rules, the open source code is extra replicated. If it doesn't work correctly, ask a question in the issue and we'll fix the problem as soon as possible.

New

Abstract

In the field of autonomous driving, online high-definition (HD) map reconstruction is crucial for planning tasks. Recent research has developed several high-performance HD map reconstruction models to meet this necessity. However, the point sequences within the instance vectors may be jittery or jagged due to prediction bias, which can impact subsequent tasks. Therefore, this paper proposes the Anti-disturbance Map (ADMap) framework. To mitigate point-order jitter, the framework consists of three modules: Multi-Scale Perception Neck, Instance Interactive Attention (IIA), and Vector Direction Difference Loss (VDDL). By exploring the point-order relationships between and within instances in a cascading manner, the model can monitor the point-order prediction process more effectively. We confirmed the validity of ADMap in both nuScenes and Argoverse2, demonstrating its excellent performance.

pipeline

paper

Main Result

nuScenes val

MethodBackbone$AP_{div}$$AP_{ped}$$AP_{bou}$mAPFPS
MapTRR5051.546.353.150.315.1
ADMapR5056.249.457.954.514.8
MapTRR50 & SECOND55.962.369.362.56.0
ADMapR50 & SECOND66.663.374.068.05.8
MapTRv2R50 & SECOND65.666.574.869.05.8
ADMapv2R50 & SECOND67.968.574.570.36.1

Argoverse2 val

MethodBackbone$AP_{div}$$AP_{ped}$$AP_{bou}$mAPFPS
MapTRR5065.556.661.861.314.8
ADMapR5068.960.364.964.714.2
MapTRv2R50 & SECOND62.972.167.167.412.0
ADMapv2R50 & SECOND72.464.568.968.713.9

Visualization results

nuScenes Visualization

nuScenes Visualization

Argoverse2 Visualization

Argoverse2 Visualization

Acknowlegement

We sincerely thank the authors of MapTR for open sourcing their methods.