Arbitrary-Scale 3D Gaussian Super-Resolution

2025-08-22

AI Summary

Existing 3D Gaussian splatting (3DGS)-based super-resolution methods are constrained by fixed integer scaling factors, compromising both computational efficiency and arbitrary-scale rendering quality; direct rendering introduces aliasing artifacts, while post-hoc upsampling degrades real-time performance. This paper proposes the first unified framework supporting arbitrary-scale (both integer and non-integer) super-resolution for 3D Gaussian point clouds. Built upon 3DGS, our end-to-end trainable approach integrates scale-aware rendering, generative prior-guided optimization, and a progressive super-resolution strategy. It ensures cross-scale structural consistency and enables high-fidelity real-time rendering with a single model: achieving 85 FPS at 1080p resolution, with a PSNR gain of +6.59 dB over baseline 3DGS. The method effectively suppresses jaggies and significantly enhances multi-scale visual fidelity.

Technology Category

• Computer Vision: 3D Computer Vision
• Machine Learning: Scalability of ML Systems
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Existing 3D Gaussian Splatting (3DGS) super-resolution methods typically perform high-resolution (HR) rendering of fixed scale factors, making them impractical for resource-limited scenarios. Directly rendering arbitrary-scale HR views with vanilla 3DGS introduces aliasing artifacts due to the lack of scale-aware rendering ability, while adding a post-processing upsampler for 3DGS complicates the framework and reduces rendering efficiency. To tackle these issues, we build an integrated framework that incorporates scale-aware rendering, generative prior-guided optimization, and progressive super-resolving to enable 3D Gaussian super-resolution of arbitrary scale factors with a single 3D model. Notably, our approach supports both integer and non-integer scale rendering to provide more flexibility. Extensive experiments demonstrate the effectiveness of our model in rendering high-quality arbitrary-scale HR views (6.59 dB PSNR gain over 3DGS) with a single model. It preserves structural consistency with LR views and across different scales, while maintaining real-time rendering speed (85 FPS at 1080p).

Problem

• Enabling arbitrary-scale 3D Gaussian super-resolution rendering
• Eliminating aliasing artifacts in vanilla 3DGS upscaling
• Maintaining real-time performance while avoiding post-processing upsamplers

Innovation

• Scale-aware rendering for arbitrary-scale super-resolution
• Generative prior-guided optimization for quality enhancement
• Progressive super-resolving with single 3D model