TEGLO: High Fidelity Canonical Texture Mapping from Single-View Images

TL;DR

TEGLO introduces a novel method for high-fidelity 3D texture mapping from single-view images, enabling detailed, editable 3D reconstructions with texture transfer capabilities without explicit 3D supervision.

Contribution

The paper presents TEGLO, a new approach that learns 3D representations from single images with high-frequency detail and editable textures, using a dense correspondence to a canonical space.

Findings

Achieves >= 74 dB PSNR at 1024^2 resolution for reconstruction.

Enables texture transfer and editing without shared mesh topology.

Provides high-quality 3D consistent novel view synthesis.

Abstract

Recent work in Neural Fields (NFs) learn 3D representations from class-specific single view image collections. However, they are unable to reconstruct the input data preserving high-frequency details. Further, these methods do not disentangle appearance from geometry and hence are not suitable for tasks such as texture transfer and editing. In this work, we propose TEGLO (Textured EG3D-GLO) for learning 3D representations from single view in-the-wild image collections for a given class of objects. We accomplish this by training a conditional Neural Radiance Field (NeRF) without any explicit 3D supervision. We equip our method with editing capabilities by creating a dense correspondence mapping to a 2D canonical space. We demonstrate that such mapping enables texture transfer and texture editing without requiring meshes with shared topology. Our key insight is that by mapping the input image pixels onto the texture space we can achieve near perfect reconstruction (>= 74 dB PSNR at 1024^2 resolution). Our formulation allows for high quality 3D consistent novel view synthesis with high-frequency details at megapixel image resolution.