Distilled-3DGS:Distilled 3D Gaussian Splatting

2025-08-19

AI Summary

To address the excessive memory and storage overhead in 3D Gaussian Splatting (3DGS) for novel-view synthesis—caused by its large number of Gaussians—this paper proposes the first knowledge distillation framework tailored for 3DGS. Our method introduces a novel multi-teacher ensemble distillation scheme, where teacher diversity is enhanced via noise injection and Dropout regularization. We further design a structural similarity loss to preserve geometric distribution consistency between teachers and student, and jointly optimize the lightweight student model using both this loss and standard rendering losses. Experiments on multiple benchmark datasets demonstrate that our approach reduces model storage by 40–60% while maintaining rendering quality competitive with state-of-the-art methods. Moreover, it significantly improves inference efficiency and deployment feasibility, enabling practical applications of 3DGS on resource-constrained platforms.

Technology Category

• Computer Vision: 3D Computer Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

Abstract

3D Gaussian Splatting (3DGS) has exhibited remarkable efficacy in novel view synthesis (NVS). However, it suffers from a significant drawback: achieving high-fidelity rendering typically necessitates a large number of 3D Gaussians, resulting in substantial memory consumption and storage requirements. To address this challenge, we propose the first knowledge distillation framework for 3DGS, featuring various teacher models, including vanilla 3DGS, noise-augmented variants, and dropout-regularized versions. The outputs of these teachers are aggregated to guide the optimization of a lightweight student model. To distill the hidden geometric structure, we propose a structural similarity loss to boost the consistency of spatial geometric distributions between the student and teacher model. Through comprehensive quantitative and qualitative evaluations across diverse datasets, the proposed Distilled-3DGS, a simple yet effective framework without bells and whistles, achieves promising rendering results in both rendering quality and storage efficiency compared to state-of-the-art methods. Project page: https://distilled3dgs.github.io . Code: https://github.com/lt-xiang/Distilled-3DGS .

Problem

• Reducing memory consumption in 3D Gaussian Splatting rendering
• Improving storage efficiency while maintaining high-fidelity rendering
• Distilling knowledge from teacher to student models for optimization

Innovation

• Knowledge distillation framework for 3DGS
• Structural similarity loss for geometric consistency
• Lightweight student model optimization