[Paper]

Runsong Zhu, Shi Qiu, Zhengzhe Liu, Ka-Hei Hui, Qianyi Wu, Pheng-Ann Heng, Chi-Wing Fu

TL;DR: Our paper presents a new and effective end-to-end lifting framework that achieves state-of-the-art (SOTA) performance for 3D Gaussian segmentation, without the need for pre-processing (e.g., video tracking) or post-processing (e.g., clustering).

Comparisons with existing works

3D point-level visualization

How to use the code.

To use Unified-Lift, please refer to the Usage guide. (currently under construction).

Citation

If you find this work useful in your research, please cite our paper:

@article{zhu2025rethinking,
  title={Rethinking End-to-End 2D to 3D Scene Segmentation in Gaussian Splatting},
  author={Zhu, Runsong and Qiu, Shi and Liu, Zhengzhe and Hui, Ka-Hei and Wu, Qianyi and Heng, Pheng-Ann and Fu, Chi-Wing},
  journal={arXiv preprint arXiv:2503.14029},
  year={2025}
}

Thanks

This code is based on Gaussian Grouping, OmniSeg3D-GS, Panoptic-Lifting. We thank the authors for releasing their code.