The paper introduces 2D Gaussian Splatting (2DGS), a novel approach for geometrically accurate radiance field reconstruction from multi-view RGB images. Unlike 3D Gaussian Splatting (3DGS), which uses 3D Gaussians that can conflict with thin surfaces and lack consistent geometry across views, 2DGS employs 2D Gaussian primitives, which are more aligned with the underlying geometry. The key contributions of 2DGS include: 1. **2D Gaussian Primitives**: These primitives are oriented elliptical disks that provide view-consistent geometry and intrinsic surface modeling. 2. **Ray-Splat Intersection**: This process ensures perspective-correct splatting, improving the accuracy of the reconstructed geometry. 3. **Depth Distortion and Normal Consistency**: These regularization terms enhance the quality of the reconstructions by addressing issues such as depth consistency and normal alignment. The method is evaluated on several datasets, including DTU, Tanks and Temples, and Mip-NerFS60, demonstrating superior performance in terms of Chamfer distance, reconstruction speed, and appearance quality compared to state-of-the-art methods. The paper also discusses the limitations of the method, such as handling semi-transparent surfaces and the trade-offs between image quality and geometry accuracy. Overall, 2DGS offers a fast and accurate solution for geometrically accurate radiance field reconstruction and novel view synthesis.The paper introduces 2D Gaussian Splatting (2DGS), a novel approach for geometrically accurate radiance field reconstruction from multi-view RGB images. Unlike 3D Gaussian Splatting (3DGS), which uses 3D Gaussians that can conflict with thin surfaces and lack consistent geometry across views, 2DGS employs 2D Gaussian primitives, which are more aligned with the underlying geometry. The key contributions of 2DGS include: 1. **2D Gaussian Primitives**: These primitives are oriented elliptical disks that provide view-consistent geometry and intrinsic surface modeling. 2. **Ray-Splat Intersection**: This process ensures perspective-correct splatting, improving the accuracy of the reconstructed geometry. 3. **Depth Distortion and Normal Consistency**: These regularization terms enhance the quality of the reconstructions by addressing issues such as depth consistency and normal alignment. The method is evaluated on several datasets, including DTU, Tanks and Temples, and Mip-NerFS60, demonstrating superior performance in terms of Chamfer distance, reconstruction speed, and appearance quality compared to state-of-the-art methods. The paper also discusses the limitations of the method, such as handling semi-transparent surfaces and the trade-offs between image quality and geometry accuracy. Overall, 2DGS offers a fast and accurate solution for geometrically accurate radiance field reconstruction and novel view synthesis.