JiSAM (CVPR 2025)

Official code for CVPR 2025 paper [JiSAM: Alleviate Labeling Burden and Corner Case Problems in Autonomous Driving via Minimal Real-World Data]. Developed based on [OpenPCDet]

Highlights

• Jittering Augmentation and Memory-based Sectorized Alignment to bridge simulation-to-real gap (JiSAM).
• JointTrainingDataset to enable training with various datasets (any number you want), supporting future research on joint training with different LiDAR datasets.
• Domain-aware Backbone follows the joint dataset to add separate input kernels for different datasets.

Getting Started

Installation

Prepare Datasets

Run example config

ToDO List

• Upload simulation dataset from CARLA and complete dataset preparation instruction.
• Example training scripts
• Explore to use SOTA generative model to create simulation dataset.

Citation

If you find this project useful in your research, please consider cite:

@InProceedings{JiSAM_CVPR_2025,
    author    = {Chen, Runjian and Shao, Wenqi and Zhang, Bo and Shi, Shaoshuai and Jiang, Li and Luo, Ping},
    title     = {JiSAM: Alleviate Labeling Burden and Corner Case Problems in Autonomous Driving via Minimal Real-World Data},
    booktitle = {Proceedings of the Computer Vision and Pattern Recognition Conference (CVPR)},
    month     = {June},
    year      = {2025},
    pages     = {6792-6801}
}

@article{JiSAM_arxiv,
  title={JiSAM: Alleviate Labeling Burden and Corner Case Problems in Autonomous Driving via Minimal Real-World Data},
  author={Chen, Runjian and Shao, Wenqi and Zhang, Bo and Shi, Shaoshuai and Jiang, Li and Luo, Ping},
  journal={arXiv preprint arXiv:2503.08422},
  year={2025}
}