Fuse and Refine
Robust and accurate HD map auto-labeling for autonomous-driving scenes.
Fuse and Refine is an HD map auto-labeling framework for static road elements in large-scale autonomous-driving scenes.
Starting from VMA as the baseline, the project builds a complete training and evaluation pipeline on self-collected data. It fuses image and LiDAR information, uses Sp-Net to improve cross-scene coarse detection robustness, projects initial predictions back to raw point-cloud space for local sampling, and uses Refine-Net to correct geometric position and shape errors.
Highlights:
- Multi-modal HD map auto-labeling.
- Image and LiDAR fusion for static road elements.
- Coarse-to-fine geometry refinement.
- Submitted to IROS as a first-author work.