[ECCV2024] InsMapper:

This is the official implementation of paper InsMapper: Exploring Inner-instance Information for Vectorized HD Mapping by Zhenhua Xu, Kwan-Yee. K. Wong, Hengshuang Zhao from the Univerisity of Hong Kong.

Project page

Update

• Jun/2024: Accepted by ECCV 2024.

• Sep/28/2023: Update the manuscript. More experiments are added, including the comparison with MapTR-V2.

• Aug/24/2023: Release more demos and the supplementary document at our new project page. Code will be released in a later stage.

• Aug/17/2023: Release paper on arxiv at paper.

Training

# find the config script in ./projects/configs/eccv
./train.sh

Testing

./test.sh

Demo

We visualize live demos of InsMapper under various scenarios, with different whether and time. Some scenes may have complicated road intersection areas. Short demos are visualized with 2Hz, while long demos have 4Hz FPS.

For more demos, please visit our new project page.

Demo 3. Short, 2Hz, day time, complicated road intersections

Demo 7. Long, 4Hz, day time, rainy, complicated road intersections

Demo 8. Long, 4Hz, night

Acknowledgement

We thank these high-quality open-sourced projects MapTR, VectorMapNet, HDMapNet, STSU, Sat2Graph, Deformable DETR.

Note

This repo is not maintained. The code should work fine but without further testing.

Citation

@inproceedings{xu2024insmapper,
  title={Insmapper: Exploring inner-instance information for vectorized hd mapping},
  author={Xu, Zhenhua and K. Wong, Kwan-Yee and Zhao, Hengshuang},
  booktitle={European Conference on Computer Vision},
  pages={296--312},
  year={2024},
  organization={Springer}
}