Advancing 3D Gaussian Splatting: Efficiency, Fidelity, and Versatility

The field of 3D Gaussian Splatting (3DGS) is witnessing significant advancements, particularly in enhancing rendering speed, fidelity, and the generation of Gaussian primitives. Recent developments focus on integrating normal vectors into the rendering pipeline to improve surface estimation and rendering quality, as seen in approaches like Normal-GS. Additionally, there is a growing emphasis on dynamic scene rendering and the use of novel encoding methods to address the limitations of traditional plane-based methods, exemplified by Grid4D. The field is also making strides in compositional 3D generation, where models like CompGS leverage 2D compositionality to initialize Gaussian parameters, ensuring consistent 3D priors and reasonable interactions among multiple entities. Furthermore, the integration of levels of detail into Gaussian avatars, as demonstrated by LoDAvatar, is balancing visual quality and computational costs, enhancing runtime frame rates. Notably, the use of Gaussian splatting for high-fidelity dynamic scene rendering and the incorporation of visual prompts in text-to-3D generation are areas of innovation that are pushing the boundaries of what is possible in this domain. These advancements collectively indicate a trend towards more efficient, high-quality, and versatile applications of 3DGS across various tasks and industries.

Sources

DiffGS: Functional Gaussian Splatting Diffusion

Normal-GS: 3D Gaussian Splatting with Normal-Involved Rendering

ODGS: 3D Scene Reconstruction from Omnidirectional Images with 3D Gaussian Splattings

CompGS: Unleashing 2D Compositionality for Compositional Text-to-3D via Dynamically Optimizing 3D Gaussians

LoDAvatar: Hierarchical Embedding and Adaptive Levels of Detail with Gaussian Splatting for Enhanced Human Avatars

Grid4D: 4D Decomposed Hash Encoding for High-fidelity Dynamic Gaussian Splatting

TV-3DG: Mastering Text-to-3D Customized Generation with Visual Prompt

ArCSEM: Artistic Colorization of SEM Images via Gaussian Splatting

GS-Blur: A 3D Scene-Based Dataset for Realistic Image Deblurring

GeoSplatting: Towards Geometry Guided Gaussian Splatting for Physically-based Inverse Rendering

URAvatar: Universal Relightable Gaussian Codec Avatars

Built with on top of