We present HS-SLAM, a framework marrying a Hybrid representation with Structural supervision for dense RGB-D SLAM.
The former leverages the complementary advantages of hash grid, tri-planes and one-blob, achieving superior scene reconstruction. For the latter, we incorporate patch rendering loss, enabling HS-SLAM to better capture structural information and perceive changes in unobserved regions. Finally, we introduce an active global BA to allocate more samples to historical observation where new regions emerge or scenes are being forgotten. This avoids re-training well-optimized scenes while eliminating camera drift and mitigating accumulative errors.
We evaluate our method on the three challenging datasets, Replica, ScanNet, and TUM. HS-SLAM demonstrates superior scene representation capabilities, effectively capturing intricate textures and achieving higher completeness in challenging areas. Furthermore, HS-SLAM highlights its effects in reducing cumulative camera pose errors, specifically when handling newly explored scenes or scenes being forgotten.
Annotations: For each scene, we provide detailed drawings, highlighting complex regions to emphasize the textures and completeness of our reconstructed scenes. HS-SLAM demonstrates superior scene representation capabilities, effectively capturing intricate textures (e.g., flowers and nightstands) and achieving higher completeness in challenging areas (e.g., chairs and walls).
Our final mesh and tracking results are available now. You can find it in the above Results button or you can download it through Google Drive.
There's a lot of excellent work that was introduced around the same time as ours.
How NeRFs and 3D Gaussian Splatting are Reshaping SLAM: a Survey provides the first comprehensive survey of SLAM progress through the lens of the latest advancements in radiance fields (NeRF and 3DGS).
Go-SLAM introduces a deep-learning-based dense visual SLAM framework that achieves real-time global optimization of poses and 3D reconstruction.
@inproceedings{Gong2025HS-SLAM,
author = {Ziren Gong and Fabio Tosi and Youmin Zhang and Stefano Mattoccia and Matteo Poggi},
title = {HS-SLAM: Hybrid Representation with Structural Supervision for Improved Dense SLAM},
booktitle = {Proceedings of the IEEE International Conference on Robotics and Automation (ICRA)},
year = {2025}
}