Improving Multi-View Reconstruction via Texture-Guided Gaussian-Mesh Joint Optimization

TL;DR

This paper introduces a unified framework for 3D object reconstruction from multi-view images that jointly optimizes geometry and appearance, leading to high-quality, editable 3D models suitable for downstream tasks.

Contribution

It proposes a novel Gaussian-guided differentiable rendering approach for simultaneous geometry and appearance optimization, bridging the gap between geometric accuracy and photorealism.

Findings

Achieves high-quality 3D reconstructions with seamless geometry and appearance.

Enables downstream editing like relighting and shape deformation.

Outperforms existing methods in reconstruction quality.

Abstract

Reconstructing real-world objects from multi-view images is essential for applications in 3D editing, AR/VR, and digital content creation. Existing methods typically prioritize either geometric accuracy (Multi-View Stereo) or photorealistic rendering (Novel View Synthesis), often decoupling geometry and appearance optimization, which hinders downstream editing tasks. This paper advocates an unified treatment on geometry and appearance optimization for seamless Gaussian-mesh joint optimization. More specifically, we propose a novel framework that simultaneously optimizes mesh geometry (vertex positions and faces) and vertex colors via Gaussian-guided mesh differentiable rendering, leveraging photometric consistency from input images and geometric regularization from normal and depth maps. The obtained high-quality 3D reconstruction can be further exploit in down-stream editing tasks, such as relighting and shape deformation. Our code will be released in https://github.com/zhejia01/TexGuided-GS2Mesh