InfiniSplat: Implicit Gaussian Decoding for Large-Baseline Monocular View Synthesis

2026-08-03

AI Summary

Existing single-image feed-forward 3D Gaussian splatting methods suffer from weak coupling between Gaussian primitives and the underlying surface due to their reliance on pixel-aligned representations, leading to poor structural consistency under large viewpoint changes. This work proposes a surface-aligned 3D Gaussian splatting framework that decouples Gaussian prediction from fixed pixel centers by leveraging geometrically guided sampling to determine 2D support points and employing a query-conditioned implicit decoder to predict Gaussian attributes. By explicitly modeling depth-induced local surface structure, the method significantly enhances structural consistency and generalization in large-baseline novel view synthesis. It achieves state-of-the-art performance across multiple datasets in cross-domain tasks and demonstrates strong zero-shot generalization from synthetic indoor scenes to complex open-world environments.

Abstract

Single-image feed-forward 3D Gaussian Splatting (3DGS) aims to directly generate a renderable 3D scene representation from one input image, avoiding the cost of multi-view capture and per-scene optimization. However, existing methods are often constrained by a pixel-aligned representation, where Gaussians are predicted from fixed image-grid locations. Such pixel-aligned primitives can produce promising nearby-view renderings, but they remain weakly coupled to underlying scene surfaces and struggle to preserve coherent structures under large viewpoint shifts. We present InfiniSplat, a feed-forward single-image 3DGS framework that moves from a pixel-aligned representation toward a surface-aligned representation. InfiniSplat constructs this representation by first using geometry-guided sampling to place 2D supports according to depth-induced local surface structure, and then applying a query-conditioned implicit decoder to predict Gaussian attributes from the image features queried at these supports.By grounding support locations in geometry while decoupling Gaussian prediction from fixed pixel centers, InfiniSplat produces Gaussian layouts that better follow scene surfaces and reduce scattered primitives caused by grid discretization.Across multiple cross-dataset NVS evaluations, InfiniSplat achieves state-of-the-art performance compared with single-image feed-forward baselines, and demonstrates zero-shot generalization from Hypersim indoor synthetic training to complex open-world scenes.Project page: https://zju3dv.github.io/InfiniSplat.

Problem

• monocular view synthesis
• 3D Gaussian Splatting
• large-baseline rendering
• surface alignment
• single-image 3D reconstruction

Innovation

• surface-aligned representation
• implicit Gaussian decoding
• geometry-guided sampling
• single-image 3D reconstruction
• zero-shot generalization