Advances in 3D Gaussian Splatting for Efficient Rendering and Reconstruction

The field of 3D Gaussian Splatting (3DGS) is rapidly advancing, with a focus on improving efficiency, accuracy, and scalability in various applications, including rendering, reconstruction, and cryo-EM. Recent developments have introduced novel frameworks, such as PowerGS, Proxy-GS, and ExGS, which jointly minimize rendering and display power, exploit proxy meshes for occlusion awareness, and achieve extreme compression ratios while preserving high rendering quality. Other notable advancements include the use of local view transformers for large-scale scene reconstruction, localized high-resolution reconstruction via on-demand Gaussian densification, and load-balanced and efficient 3DGS for large-scale scenes. Noteworthy papers include PowerGS, which achieves up to 86% total power reduction, and ExGS, which enables over 100x compression while preserving fidelity. Additionally, papers like Proxy-GS and LoBE-GS demonstrate significant improvements in rendering speed and training efficiency, while GEM and GaussianLens showcase the potential of 3DGS in cryo-EM reconstruction and localized high-resolution reconstruction, respectively.

Sources

PowerGS: Display-Rendering Power Co-Optimization for Neural Rendering in Power-Constrained XR Systems

Proxy-GS: Efficient 3D Gaussian Splatting via Proxy Mesh

ExGS: Extreme 3D Gaussian Compression with Diffusion Priors

LVT: Large-Scale Scene Reconstruction via Local View Transformers

GEM: 3D Gaussian Splatting for Efficient and Accurate Cryo-EM Reconstruction

GaussianLens: Localized High-Resolution Reconstruction via On-Demand Gaussian Densification

LOBE-GS: Load-Balanced and Efficient 3D Gaussian Splatting for Large-Scale Scene Reconstruction

ROI-GS: Interest-based Local Quality 3D Gaussian Splatting

Built with on top of