3DEnhancer : Consistent Multi-View Diffusion for 3D Enhancement

Yihang Luo, Shangchen ZhouYushi LanXingang Pan, Chen Change LoyS-Lab, Nanyang Technological University https://yihangluo.com/projects/3DEnhancer · 2024

Read 3DEnhancer : Consistent Multi-View Diffusion for 3D Enhancement by Yihang Luo, Shangchen ZhouYushi LanXingang Pan, Chen Change LoyS-Lab, Nanyang Technological University https://yihangluo.com/projects/3DEnhancer (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-18565.txt.

Description

Yihang Luo and co-authors present 3DEnhancer, a 2024 system that upgrades rough 3D models by polishing all their views at once. Neural rendering is held back by scarce high-quality 3D data and by diffusion models that disagree between viewpoints, so generated models stay blurry and inconsistent.

A pose-aware encoder feeds a multi-view diffusion denoiser that refines low-quality renders while an attention module with epipolar geometry keeps views in agreement. Data augmentation widens the training diet. Unlike video-based rivals, the approach enhances any set of angles seamlessly. Evaluations show clear wins in both multi-view enhancement and per-object 3D refinement tasks.

Catalog record

CREATORYihang Luo, Shangchen ZhouYushi LanXingang Pan, Chen Change LoyS-Lab, Nanyang Technological University https://yihangluo.com/projects/3DEnhancer
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE55 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-18565

Browse the full library · About the Tower of Babel