Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces G$^2$SR, a novel approach for few-view surface reconstruction that combines a lightweight neural network with an analytic backend to efficiently derive 3D Gaussian splats from 2D image correspondences. By leveraging multi-view geometry, G$^2$SR significantly reduces memory usage and computational demands compared to traditional end-to-end methods, which often rely on large Transformer networks. The method achieves competitive geometric accuracy on datasets like ScanNet and DTU, processing 69-89 reconstructions per second while maintaining a memory footprint of only 203 MB, making it suitable for real-time applications on mobile platforms.
G$^2$SR achieves state-of-the-art geometric accuracy in surface reconstruction while operating at a fraction of the memory and computational cost of existing methods.
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.