MemGS: Memory-Efficient Gaussian Splatting for Real-Time SLAM
🤖 AI Summary
To address the trade-off between reconstruction quality and real-time performance in 3D Gaussian splatting on memory-constrained embedded platforms (e.g., micro aerial vehicles), this paper proposes a memory-efficient Gaussian lattice rendering method. The approach operates by clustering geometrically similar Gaussian primitives within a voxelized space, significantly reducing GPU memory footprint, and introduces a Patch-Grid point cloud sampling strategy to enhance initialization accuracy and reconstruction completeness. By preserving GPU computational efficiency while drastically lowering memory consumption, the method achieves balanced improvements in rendering fidelity, geometric completeness, and frame rate. Extensive evaluation on public benchmarks demonstrates consistent gains across all three metrics, validating its effectiveness for resource-limited SLAM systems. This work provides a deployable 3D Gaussian reconstruction solution tailored for embedded platforms with stringent memory budgets.
Application Category
📝 Abstract
Recent advancements in 3D Gaussian Splatting (3DGS) have made a significant impact on rendering and reconstruction techniques. Current research predominantly focuses on improving rendering performance and reconstruction quality using high-performance desktop GPUs, largely overlooking applications for embedded platforms like micro air vehicles (MAVs). These devices, with their limited computational resources and memory, often face a trade-off between system performance and reconstruction quality. In this paper, we improve existing methods in terms of GPU memory usage while enhancing rendering quality. Specifically, to address redundant 3D Gaussian primitives in SLAM, we propose merging them in voxel space based on geometric similarity. This reduces GPU memory usage without impacting system runtime performance. Furthermore, rendering quality is improved by initializing 3D Gaussian primitives via Patch-Grid (PG) point sampling, enabling more accurate modeling of the entire scene. Quantitative and qualitative evaluations on publicly available datasets demonstrate the effectiveness of our improvements.
Problem
Research questions and friction points this paper is trying to address.
Reduces GPU memory usage for embedded platforms
Merges redundant 3D Gaussian primitives in SLAM
Improves rendering quality via Patch-Grid sampling
Innovation
Methods, ideas, or system contributions that make the work stand out.
Merging redundant Gaussians in voxel space
Initializing Gaussians via Patch-Grid sampling
Reducing GPU memory while enhancing rendering
🔎 Similar Papers
No similar papers found.
