Gaussian Splatting for Immersive Video and 3D Avatars

The field of computer vision and graphics is moving towards the development of more realistic and immersive video and 3D avatars. Recent research has focused on using Gaussian splatting techniques to achieve high-quality rendering and reconstruction. This has led to advancements in areas such as dynamic scene reconstruction, neural video compression, and 3D stylized head avatars. The use of Gaussian splatting has enabled the creation of more detailed and realistic models, and has improved the efficiency of rendering and compression algorithms. Noteworthy papers include TeGA, which presents a high-detail 3D head avatar model, and GIFStream, which introduces a novel 4D Gaussian representation for immersive video. ADC-GS is also notable for its compact and efficient representation for dynamic scene reconstruction. Additionally, Neural Video Compression using 2D Gaussian Splatting and ToonifyGB demonstrate the potential of Gaussian splatting for neural video compression and 3D stylized head avatars.

Sources

TeGA: Texture Space Gaussian Avatars for High-Resolution Dynamic Head Modeling

GIFStream: 4D Gaussian-based Immersive Video with Feature Stream

ADC-GS: Anchor-Driven Deformable and Compressed Gaussian Splatting for Dynamic Scene Reconstruction

Neural Video Compression using 2D Gaussian Splatting

ToonifyGB: StyleGAN-based Gaussian Blendshapes for 3D Stylized Head Avatars

Built with on top of