GoMAvatar: Efficient Animatable Human Modeling from Monocular Video Using Gaussians-on-Mesh

GoMAvatar, a groundbreaking approach to animatable human modeling, has been introduced, offering real-time performance, memory efficiency, and high-quality results. This innovative method requires only a single monocular video to generate a digital avatar that can be re-articulated in new poses and rendered from novel viewpoints in real-time, seamlessly integrating with rasterization-based graphics pipelines.

The core of GoMAvatar lies in its unique Gaussians-on-Mesh representation, a hybrid 3D model that combines the rendering quality and speed of Gaussian splatting with the geometry modeling and compatibility of deformable meshes. This powerful combination enables GoMAvatar to deliver exceptional results while maintaining optimal performance.

Extensive testing on ZJU-MoCap data and various YouTube videos has demonstrated Go MAvatar’s superior performance compared to existing monocular human modeling algorithms. It matches or surpasses current methods in rendering quality while significantly outperforming them in computational efficiency, achieving an impressive 43 frames per second. Additionally, GoMAvatar is highly memory-efficient, requiring only 3.63 MB per subject.

With its ability to create highly realistic, animatable human models from a single monocular video input, GoMAvatar has the potential to revolutionize various applications, including virtual reality, gaming, and film production. Its real-time performance, memory efficiency, and seamless integration with existing graphics pipelines make it an attractive solution for developers and content creators seeking to enhance their projects with lifelike human avatars.

The introduction of GoMAvatar marks a significant advancement in animatable human modeling, offering a powerful, efficient, and accessible tool for creating high-quality digital avatars from monocular video input.