GS-Pose: Generalizable Segmentation-based 6D
Object Pose Estimation with 3D Gaussian Splatting

Dingding Cai
Tampere University
Janne Heikkilä
University of Oulu
Esa Rahtu
Tampere University

International Conference on 3D Vision 2025

Overview of GS-Pose

(1). Create the reference database for a newly added object

(2). Perform pose estimation for the added object

(Additional) (3). Perform pose tracking for the added object

Comparison with OnePose++

All results are from single-frame pose estimation without tracking. Ours achieves relatively more stable and accurate results.

Pose Tracking Results

We compare the pose estimation results with/without pose tracking.

Quantitative Results

Citation

@InProceedings{cai_2024_GSPose,
    author    = {Cai, Dingding and Heikkil\"a, Janne and Rahtu, Esa},
    title     = {GS-Pose: Generalizable Segmentation-based 6D Object Pose Estimation with 3D Gaussian Splatting},
    journal   = {arXiv preprint arXiv:2403.10683v2},
    year      = {2024},
}

Acknowledgements

This webpage template is based on link.