HDMapNet

HDMapNet

HDMapNet: An Online HD Map Construction and Evaluation Framework

ICRA 2022, CVPR 2021 Workshop best paper nominee

Qi Li, Yue Wang, Yilun Wang, Hang Zhao

[Paper] [Project Page] [5-min video]

Abstract: Estimating local semantics from sensory inputs is a central component for high-definition map constructions in autonomous driving. However, traditional pipelines require a vast amount of human efforts and resources in annotating and maintaining the semantics in the map, which limits its scalability. In this paper, we introduce the problem of local semantic map learning, which dynamically constructs the vectorized semantics based on onboard sensor observations. Meanwhile, we introduce a local semantic map learning method, dubbed HDMapNet. HDMapNet encodes image features from surrounding cameras and/or point clouds from LiDAR, and predicts vectorized map elements in the bird's-eye view. We benchmark HDMapNet on nuScenes dataset and show that in all settings, it performs better than baseline methods. Of note, our fusion-based HDMapNet outperforms existing methods by more than 50% in all metrics. In addition, we develop semantic-level and instance-level metrics to evaluate the map learning performance. Finally, we showcase our method is capable of predicting a locally consistent map. By introducing the method and metrics, we invite the community to study this novel map learning problem. Code and evaluation kit will be released to facilitate future development.

Questions/Requests: Please file an issue or email [email protected].

Preparation

1. Download nuScenes dataset and put it to dataset/ folder.

2. Install dependencies by running

pip install -r requirement.txt

1. Install pytorch from https://pytorch.org/get-started/locally/

2. Install pytorch scatter from https://pytorch-geometric.readthedocs.io/en/latest/notes/installation.html

Label

Run python vis_label.py for demo of vectorized labels. The visualizations are in dataset/nuScenes/samples/GT.

Training

Run python train.py --instance_seg --direction_pred --version [v1.0-trainval, v1.0-mini] --logdir [output place].

Evaluation

Before running the evaluation code, you should get the submission.json file first, which can be generated by the following command.

python export_gt_to_json.py

Run python evaluate.py --result_path [submission file path] for evaluation. The script accepts vectorized or rasterized maps as input. For vectorized map, We firstly rasterize the vectors to map to do evaluation. For rasterized map, you should make sure the line width=1.

Below is the format for vectorized submission:

…

For rasterized submission, the format is:

…

Run python export_gt_to_json.py to get a demo of vectorized submission. Run python export_gt_to_json.py --raster for rasterized submission.

Run python export_pred_to_json.py --modelf [checkpoint] to get submission file for trained model.

Citation

If you found this paper or codebase useful, please cite our paper:

@misc{li2021hdmapnet,
      title={HDMapNet: An Online HD Map Construction and Evaluation Framework}, 
      author={Qi Li and Yue Wang and Yilun Wang and Hang Zhao},
      year={2021},
      eprint={2107.06307},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}