Reproduce “Structure-aware Editable Morphable Model for 3D Facial Detail Animation and Manipulation” 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
Jingwang Ling, Zhibo Wang, Ming Lu, Quan Wang, Chen Qian, Feng Xu
The Structure-aware Editable Morphable Model (SEMM) provides a method for representing, animating, and editing fine-grained facial details such as wrinkles and creases. The approach leverages distance field representations of wrinkle structures and specialized transformation modules to enable intuitive control over expression-driven and age-related detail changes, resulting in more realistic facial animations than prior methods.
Key insights
- Representing facial wrinkle lines as distance fields, rather than direct displacement maps, enables implicit correspondence and better preservation of structural details across different faces and expressions.
- Specialized transformation modules, trained within an autoencoder framework, can effectively learn to animate and edit facial details based on external parameters like expression blendshape weights and age values.
- A multi-task discriminator and various loss functions (reconstruction, adversarial, cycle consistency, feature matching) are crucial for generating realistic details and ensuring meaningful transformations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.