Reproduce “Skull-to-Face: Anatomy-Guided 3D Facial Reconstruction and Editing” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yongqing Liang, Congyi Zhang, Junli Zhao, Wenping Wang, Xin Li
An end-to-end pipeline generates realistic, textured 3D faces from skull scans by combining biological profile-guided initial face generation with anatomy-guided adaptation based on learned tissue depth distributions. The method achieves a mean reconstruction error of 1.526% on a test dataset, outperforming previous methods and allowing interactive exploration of plausible facial appearances.
Key insights
- Modeling tissue depth distributions with Principal Component Analysis (PCA) allows capturing significant variations in facial fullness (global) and localized features (regional), enabling the generation of diverse yet anatomically valid faces from a single skull.
- Combining powerful generative models like Stable Diffusion XL for realistic texture and initial plausible shapes with a robust, anatomy-guided optimization ensures both visual realism and precise geometric conformity to the skull.
- The face adaptation mechanism is robust enough to converge diverse initial faces to skull-conforming geometries while retaining subtle inherited textural details, indicating a blend of anatomical accuracy and visual richness.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.