Importance-Driven 3D Gaussian SLAM for Efficient Mapping and Communication-Aware Sharing

Zhaolin Yang, Huilin Yin, Linchuan Zhang, Mingyu Liu, Alois Knoll

Keywords: 3D Gaussian Splatting, 3D Scene Reconstruction, SLAM, Map Compression, Bandwidth-Constrained Communication

TL;DR: We propose PWAS, a unified importance metric for 3D Gaussian Splatting SLAM that enables efficient map compression and bandwidth-aware transmission.

Abstract:

3D Gaussian Splatting (3DGS) enables efficient dense Simultaneous Localization and Mapping (SLAM) but typically produces highly redundant maps, leading to increased computational and communication costs in resource-constrained settings. In this work, we propose a unified importance-driven Gaussian selection framework that addresses both map redundancy and multi-agent communication efficiency.
We introduce Proximity-Weighted Accumulation Score (PWAS), a multi-view Gaussian importance metric that quantifies the structural contribution and cross-view consistency of each Gaussian primitive. Based on this metric, we design an importance-aware pruning strategy that removes redundant Gaussians while preserving reconstruction accuracy. Furthermore, we extend the same importance metric to cooperative SLAM and propose an selective sharing mechanism under limited communication budgets. Instead of transmitting full maps, agents exchange only the most informative Gaussians. Experiments on Replica and TUM RGB-D demonstrate that our method significantly reduces map size and communication overhead while maintaining competitive tracking and rendering performance, and consistently outperforms random sharing under identical bandwidth constraints.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 5