UniCross3D is an arXiv paper proposing a diffusion framework that generates 3D content from a single RGB image, aiming for more consistent geometry and appearance.
Two additions: a cross-view latent regularization that penalizes latent variance to keep geometry consistent across synthesized viewpoints, and a cross-domain mutual information objective that aligns synthesized color and normal maps.
Claims: the authors say it improves view consistency and semantic alignment over state-of-the-art methods and gives higher-fidelity reconstructions, especially with challenging textures and ambiguous viewpoints.
Missing: the abstract gives no benchmark numbers, so the size of the improvement is unknown from this page alone. It is listed as an Eurographics 2026 paper in Computer Graphics Forum.
Sign in to comment.
TL;DR of the paper, generated automatically.
UniCross3D is an arXiv paper proposing a diffusion framework that generates 3D content from a single RGB image, aiming for more consistent geometry and appearance.