GMIR is a strategy to retrieve samples from old domain data that help alleviate forgetting while finetuning on new domain. This repository is based off [OpenPCDet].

Gradient-based Maximally Interfered Retrieval for Domain Adaptive 3D Object Detection
Barza Nisar, Vishal Kanna Anand, Steven L. Waslander
[Paper]

Please refer to INSTALL.md for the installation of GMIR.

Please refer to GETTING_STARTED.md to learn more usage about this project.

GMIR is released under the Apache 2.0 license.

We would like to thank the authors of OpenPCDet for their open source release of their codebase.

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

@inproceedings{GMIR,
    title={{Gradient-based Maximally Interfered Retrieval for Domain Adaptive 3D Object Detection}},
    author={Nisar, Barza and
            Annand, Vishal Kanna and
            Waslander, Steven L.},
    booktitle={Conference on Computer and Robot Vision (CRV)},
    year={2023}
}