CVPR Findings2026
Point2Gaussian: Point-Cloud-to-Gaussian Conversion for Efficient 3D Scene Rendering
Powei Liao, Jiro Abe, Kazumine Ogura
Abstract
Point clouds are widely used for 3D scene representation across various applications. However, accurate representation requires dense point clouds, leading to large data sizes that hinder real-time rendering on portable devices. Moreover, at close viewing distances, point cloud renderings appear as sparse points and thus visually differ from the corresponding real-world scene. To address these limitations, we propose Point2Gaussian, a method that converts point clouds into 3D Gaussian representations for efficient and high-fidelity rendering. Our approach renders images from point clouds using virtual cameras and then trains 3D Gaussians on these images. Experiments on a novel dataset captured with a high-accuracy laser scanner demonstrate that Point2Gaussian outperforms existing point cloud rendering methods in visual quality metrics (PSNR, SSIM, LPIPS), compact data size (93% reduction), and rendering speed (FPS), while maintaining geometric accuracy. The method enables real-time rendering of largescale 3D scenes on resource-constrained devices, making it highly suitable for industrial applications.
Ask Lune
Know this paper in minutes, not an afternoon.
Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.
Lineage
- A Hierarchical 3D Gaussian Representation for Real-Time Rendering of Very Large Datasets cites 3D Gaussian Splatting for Real-Time Radiance Field Rendering
- A Hierarchical 3D Gaussian Representation for Real-Time Rendering of Very Large Datasets cites Mip-Splatting: Alias-Free 3D Gaussian Splatting
- Mip-Splatting: Alias-Free 3D Gaussian Splatting cites 3D Gaussian Splatting for Real-Time Radiance Field Rendering
- Mip-Splatting: Alias-Free 3D Gaussian Splatting cites Mip-NeRF 360: Unbounded Anti-Aliased Neural Radiance Fields
Builds on10
- 3D Gaussian Splatting for Real-Time Radiance Field RenderingBernhard Kerbl, Georgios Kopanas, Thomas Leimkühler, George DrettakisSIGGRAPH 2023 · 5,687 citations
- Mip-NeRF 360: Unbounded Anti-Aliased Neural Radiance FieldsJonathan T. Barron, Ben Mildenhall, Dor Verbin, Pratul P. Srinivasan et al.CVPR 2022 · 1,603 citations
- 2D Gaussian Splatting for Geometrically Accurate Radiance FieldsBinbin Huang, Zehao Yu, Anpei Chen, Andreas Geiger et al.SIGGRAPH 2024 · 660 citations
- Mip-Splatting: Alias-Free 3D Gaussian SplattingZehao Yu, Anpei Chen, Binbin Huang, Torsten Sattler et al.CVPR 2024 · 360 citations
- A Hierarchical 3D Gaussian Representation for Real-Time Rendering of Very Large DatasetsBernhard Kerbl, Andreas Meuleman, Georgios Kopanas, Michael Wimmer et al.SIGGRAPH 2024 · 180 citations
Related papers
- Lightweight Predictive 3D Gaussian SplatsJunli Cao, Vidit Goel, Chaoyang Wang, Anil Kag et al.ICLR 2025
- LightGaussian: Unbounded 3D Gaussian Compression with 15x Reduction and 200+ FPSZhiwen Fan, Kevin Wang, Kairun Wen, Zehao Zhu et al.NeurIPS 2024 · 681 citations
- Point2Pix: Photo-Realistic Point Cloud Rendering via Neural Radiance FieldsTao Hu, Xiaogang Xu, Shu Liu, Jiaya JiaCVPR 2023
- Compressed 3D Gaussian Splatting for Accelerated Novel View SynthesisSimon Niedermayr, Josef Stumpfegger, Rüdiger WestermannCVPR 2024 · 138 citations
- GAP: Gaussianize Any Point Clouds with Text GuidanceWeiqi Zhang, Junsheng Zhou, Haotian Geng, Wenyuan Zhang et al.ICCV 2025 · 2 citations
