G$^2$SR: Geometric Methods for Fast and Memory-Efficient Gaussian-based Surface Reconstruction

2026-07-15

AI Summary

This work proposes an efficient and lightweight few-view RGB surface reconstruction method tailored for mobile platforms, balancing speed, accuracy, and low memory consumption. The approach decomposes reconstruction into well-posed geometric subtasks by integrating a lightweight neural front-end with an analytical geometric back-end: it first detects and tracks 2D Gaussian points in the image plane and then triangulates them via multi-view geometry to produce a metric-scale 3D Gaussian splat representation. This design circumvents the high computational overhead and generalization limitations of end-to-end networks. Evaluated on ScanNet, Replica, and DTU datasets, the method achieves or surpasses state-of-the-art geometric accuracy while reconstructing at 69–89 frames per second from only 2–3 input views, with a GPU memory footprint of merely 203 MB—5 to 107 times lower than existing approaches.

Abstract

Few-view surface reconstruction recovers the visible surfaces of a scene from a few posed RGB images, providing the 3D models that robots need to explore and interact online. On mobile platforms, the reconstruction must be fast and geometrically accurate while keeping a small memory footprint to ensure safe and efficient operation. 3D Gaussian Splatting (3DGS) offers a high-fidelity scene representation, but building it from a few views is ill-posed, as many distinct surfaces reproduce the same images, making traditional photometric methods prone to "floater" artifacts. End-to-end methods resolve the ambiguity by regressing splats with large, usually Transformer-based, networks that require heavy compute and memory while generalizing poorly to new scenes. We propose G2SR, which exploits a well-posed core of the task: given cross-view 2D splat correspondences, 3D splats follow analytically from multi-view geometry. G2SR employs a lightweight neural frontend to detect and track 2D Gaussian splats on the image plane and an analytic backend to triangulate each into a metric-scale 3D splat. On ScanNet, Replica, and DTU, G2SR matches or exceeds the geometric accuracy of state-of-the-art end-to-end methods while running at 69-89 reconstructions per second within 203 MB of GPU memory (5-107x less) for 2- and 3-view inputs at 384 x 512 resolution, offering a practical path to online Gaussian-based surface reconstruction.

Problem

• Few-view surface reconstruction
• 3D Gaussian Splatting
• floater artifacts
• memory efficiency
• geometric accuracy

Innovation

• Gaussian Splatting
• Few-view Reconstruction
• Multi-view Geometry
• Memory Efficiency
• Real-time Surface Reconstruction