Abstract
The recent 3D Gaussian splatting (3D-GS) has shown remarkable renderingfidelity and efficiency compared to NeRF-based neural scene representations.While demonstrating the potential for real-time rendering, 3D-GS encountersrendering bottlenecks in large scenes with complex details due to an excessivenumber of Gaussian primitives located within the viewing frustum. Thislimitation is particularly noticeable in zoom-out views and can lead toinconsistent rendering speeds in scenes with varying details. Moreover, itoften struggles to capture the corresponding level of details at differentscales with its heuristic density control operation. Inspired by theLevel-of-Detail (LOD) techniques, we introduce Octree-GS, featuring anLOD-structured 3D Gaussian approach supporting level-of-detail decompositionfor scene representation that contributes to the final rendering results. Ourmodel dynamically selects the appropriate level from the set ofmulti-resolution anchor points, ensuring consistent rendering performance withadaptive LOD adjustments while maintaining high-fidelity rendering results.