Temel Kavramlar
Proposing a text-driven layered 3D human generation framework based on a physically-decoupled semantic-aware diffusion model.
Özet
The content introduces the HumanCoser framework for generating layered 3D humans guided by text prompts.
It addresses the limitations of existing methods in generating high-quality layered 3D humans with consistent body structures.
The proposed method focuses on decoupling body and clothing generation, ensuring structural consistency and flexibility in editing.
Key components include a semantic-confidence strategy for clothing, SMPL-driven implicit field deformation network, and normal prediction network.
Experimental results demonstrate the effectiveness of the proposed method in generating realistic 3D humans with consistent body structures.
İstatistikler
arXiv:2312.05804v2 [cs.CV] 19 Mar 2024
Alıntılar
"Our main contributions are summarized as follows:"
"We propose a layered 3D human generation framework with a semantic-aware diffusion model."