Two-Stage Gaussian Splatting Optimization for Outdoor Scene Reconstruction

2025-10-10

AI Summary

Outdoor scene reconstruction faces a fundamental trade-off: foreground regions exhibit rich texture but distant background elements (e.g., sky) suffer from low detail, non-uniform illumination, and geometric ambiguity. To address this, we propose a two-stage Gaussian splatting optimization framework that explicitly decouples background and foreground modeling for the first time. The background is initialized via spherical shell constraints and regularized using tangent-plane smoothness, enabling automatic, label-free environment map estimation. The foreground is reconstructed under SfM guidance and jointly optimized with the background. A stage-wise rendering loss design mitigates optimization conflicts between near and far regions. Evaluated on multiple outdoor datasets, our method significantly suppresses background artifacts, improves perceptual realism in novel-view synthesis, and yields higher-fidelity environment maps—outperforming state-of-the-art approaches.

Technology Category

• Search and Optimization: Non-convex Optimization
• Planning, Routing, and Scheduling: Activity and Plan Recognition
• Computer Vision: Learning & Optimization for CV

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Outdoor scene reconstruction remains challenging due to the stark contrast between well-textured, nearby regions and distant backgrounds dominated by low detail, uneven illumination, and sky effects. We introduce a two-stage Gaussian Splatting framework that explicitly separates and optimizes these regions, yielding higher-fidelity novel view synthesis. In stage one, background primitives are initialized within a spherical shell and optimized using a loss that combines a background-only photometric term with two geometric regularizers: one constraining Gaussians to remain inside the shell, and another aligning them with local tangential planes. In stage two, foreground Gaussians are initialized from a Structure-from-Motion reconstruction, added and refined using the standard rendering loss, while the background set remains fixed but contributes to the final image formation. Experiments on diverse outdoor datasets show that our method reduces background artifacts and improves perceptual quality compared to state-of-the-art baselines. Moreover, the explicit background separation enables automatic, object-free environment map estimation, opening new possibilities for photorealistic outdoor rendering and mixed-reality applications.

Problem

• Addresses outdoor scene reconstruction challenges from stark contrast regions
• Separates and optimizes foreground and background using Gaussian Splatting
• Reduces background artifacts and enables environment map estimation

Innovation

• Two-stage Gaussian Splatting separates foreground and background optimization
• Background uses spherical shell constraints with geometric regularizers
• Foreground refines SfM reconstruction while fixing background parameters