Faster and Better 3D Splatting via Group Training

2024-12-10

arXiv.org

AI Summary

To address the high computational cost and slow convergence of 3D Gaussian Splatting (3DGS) caused by its large number of Gaussian primitives, this paper proposes a grouped collaborative training paradigm. Specifically, Gaussian primitives are dynamically clustered into groups, within which gradients are shared and updated cooperatively, augmented by adaptive sampling scheduling. This approach innovatively achieves efficient parameter optimization while preserving the geometric and radiometric fidelity of the scene representation. Crucially, it requires no modification to the underlying rendering pipeline and is plug-and-play compatible with standard 3DGS, Mip-Splatting, and other mainstream frameworks. Extensive experiments across multiple scenes demonstrate up to a 30% acceleration in training convergence, accompanied by consistent improvements in PSNR and SSIM. Moreover, the method significantly enhances reconstruction fidelity and optimization stability.

Technology Category

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

Application Category

• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

Abstract

3D Gaussian Splatting (3DGS) has emerged as a powerful technique for novel view synthesis, demonstrating remarkable capability in high-fidelity scene reconstruction through its Gaussian primitive representations. However, the computational overhead induced by the massive number of primitives poses a significant bottleneck to training efficiency. To overcome this challenge, we propose Group Training, a simple yet effective strategy that organizes Gaussian primitives into manageable groups, optimizing training efficiency and improving rendering quality. This approach shows universal compatibility with existing 3DGS frameworks, including vanilla 3DGS and Mip-Splatting, consistently achieving accelerated training while maintaining superior synthesis quality. Extensive experiments reveal that our straightforward Group Training strategy achieves up to 30% faster convergence and improved rendering quality across diverse scenarios.

Problem

• Reducing computational overhead in 3D Gaussian Splatting training
• Improving rendering quality via optimized primitive group management
• Accelerating convergence while maintaining synthesis fidelity

Innovation

• Group Training organizes Gaussian primitives efficiently
• Compatible with vanilla 3DGS and Mip-Splatting frameworks
• Achieves 30% faster convergence with better quality