Generalizable Human Gaussians for Sparse View Synthesis

TL;DR

This paper introduces a novel approach for generalizing human 3D models from sparse views by learning human Gaussians in UV space, enabling photorealistic rendering of new subjects with limited data.

Contribution

It proposes a new method to learn generalizable human Gaussians using UV space regression and a multi-scaffold, improving view synthesis from sparse views.

Findings

Outperforms recent methods in within-dataset generalization.

Achieves better cross-dataset generalization.

Enables photorealistic rendering from limited views.

Abstract

Recent progress in neural rendering has brought forth pioneering methods, such as NeRF and Gaussian Splatting, which revolutionize view rendering across various domains like AR/VR, gaming, and content creation. While these methods excel at interpolating {\em within the training data}, the challenge of generalizing to new scenes and objects from very sparse views persists. Specifically, modeling 3D humans from sparse views presents formidable hurdles due to the inherent complexity of human geometry, resulting in inaccurate reconstructions of geometry and textures. To tackle this challenge, this paper leverages recent advancements in Gaussian Splatting and introduces a new method to learn generalizable human Gaussians that allows photorealistic and accurate view-rendering of a new human subject from a limited set of sparse views in a feed-forward manner. A pivotal innovation of our approach involves reformulating the learning of 3D Gaussian parameters into a regression process defined on the 2D UV space of a human template, which allows leveraging the strong geometry prior and the advantages of 2D convolutions. In addition, a multi-scaffold is proposed to effectively represent the offset details. Our method outperforms recent methods on both within-dataset generalization as well as cross-dataset generalization settings.