MGMap: Mask-Guided Learning for Online Vectorized HD Map Construction

Xiaolu Liu, Song Wang, Wentong Li, Ruizi Yang, Junbo Chen, Jianke Zhu

[Paper] (arXiv). CVPR2024

News

• [2024/4/13]: We release the code and checkpoint for camera modality.

Video Demo

Introduction

We propose MGMap, a mask-guided approach that effectively highlights the informative regions and achieves precise map element localization by introducing the learned masks. Specifically, MGMap employs learned masks based on the enhanced multi-scale BEV features from two perspectives. At the instance level, we propose the Mask-activated instance (MAI) decoder, which incorporates global instance and structural information into instance queries by the activation of instance masks. At the point level, a novel position-guided mask patch refinement (PG-MPR) module is designed to refine point locations from a finer-grained perspective, enabling the extraction of point-specific patch information. Compared to the baselines, our proposed MGMap achieves a notable promotion of around 10 mAP for different input modalities. Extensive experiments also demonstrate that our approach showcases strong robustness and generalization capabilities.

TODO

• Release the code.

• Add configs for LiDAR and fusion modalities.

• Release pre-trained models.

Getting Started

• Installation
• Prepare Dataset
• Train and Eval

Quantitative Results

nuScenes dataset

Model	Modality	Backbone	Epoch	mAP	FPS	Config	Download
MGMap	Camera	R50	30	61.4	11.6	config	model
MGMap	Lidar	Second	24	67.9	5.5	config	model
MGMap	Camera&Lidar	R50&Sec	24	71.7	4.8	config	model

Acknowledgements

MGMap is based on mmdetection3d. It is also greatly inspired by the following outstanding contributions to the open-source community: BEVFormer, HDMapNet, MapTR, SparseInst.

Citation

If the paper and code help your research, please kindly cite:

@misc{liu2024mgmap,
      title={MGMap: Mask-Guided Learning for Online Vectorized HD Map Construction}, 
      author={Xiaolu Liu and Song Wang and Wentong Li and Ruizi Yang and Junbo Chen and Jianke Zhu},
      year={2024},
      eprint={2404.00876},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

​