Nix and Fix: Targeting 1000x Compression of 3D Gaussian Splatting with Diffusion Models

2026-02-04

AI Summary

Although 3D Gaussian Splatting (3DGS) enables real-time rendering, it suffers from severe visual artifacts under extremely high compression ratios—such as 0.1 MB—limiting its applicability in scenarios like immersive communication. To address this, this work proposes NiFi, a one-step distillation method based on diffusion models that introduces, for the first time, an artifact-aware mechanism into the post-compression reconstruction of 3DGS. NiFi successfully recovers high-quality novel views even at compression ratios as high as 1000×, significantly outperforming existing compression approaches. By doing so, it overcomes the visual fidelity bottleneck at low bitrates while maintaining perceptual quality comparable to that of the original 3DGS representation.

Technology Category

• Computer Vision: 3D Computer Vision
• Machine Learning: Learning on the Edge & Model Compression
• Data Mining & Knowledge Management: Data Compression

Application Category

• 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
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

Abstract

3D Gaussian Splatting (3DGS) revolutionized novel view rendering. Instead of inferring from dense spatial points, as implicit representations do, 3DGS uses sparse Gaussians. This enables real-time performance but increases space requirements, hindering applications such as immersive communication. 3DGS compression emerged as a field aimed at alleviating this issue. While impressive progress has been made, at low rates, compression introduces artifacts that degrade visual quality significantly. We introduce NiFi, a method for extreme 3DGS compression through restoration via artifact-aware, diffusion-based one-step distillation. We show that our method achieves state-of-the-art perceptual quality at extremely low rates, down to 0.1 MB, and towards 1000x rate improvement over 3DGS at comparable perceptual performance. The code will be open-sourced upon acceptance.

Problem

• 3D Gaussian Splatting
• compression
• artifacts
• low bitrate
• visual quality

Innovation

• 3D Gaussian Splatting
• diffusion models
• extreme compression
• artifact-aware restoration
• one-step distillation