Related Experiment Video
Updated: Nov 5, 2025
11:57
Measuring Spatially- and Directionally-varying Light Scattering from Biological Material
Published on: May 20, 2013
13.7K
Hybrid Face Reflectance, Illumination, and Shape From a Single Image
Summary
We introduce HyFRIS-Net, a novel method for estimating face shape, reflectance, and illumination from a single image. This approach achieves realistic albedo and shape recovery, outperforming existing methods.
Area of Science:
- Computer Vision
- Computer Graphics
- Machine Learning
Background:
- Accurate face modeling from single images is challenging due to complex illumination and reflectance variations.
- Existing methods often struggle with occlusions and disentangling intrinsic face properties.
Purpose of the Study:
- To develop a unified framework for jointly estimating hybrid reflectance and illumination models and refined face shape from a single unconstrained image.
- To achieve occlusion-free face albedo recovery with disambiguated color.
Main Methods:
- Proposed HyFRIS-Net (Hybrid Reflectance and Illumination Network) for joint estimation.
- Utilized a hybrid representation for reflectance and illumination modeling in parametric and non-parametric spaces.
- Enforced reflectance consistency and face identity constraints during training.
- Employed a self-evolving training strategy for general applicability.
Main Results:
- Recovered occlusion-free face albedo with disambiguated color.
- Demonstrated superior performance in modeling photo-realistic face albedo, illumination, and shape compared to state-of-the-art methods.
- Achieved general applicability on real-world data through self-evolving training.
Conclusions:
- HyFRIS-Net effectively models photometric face appearance by jointly estimating reflectance, illumination, and shape.
- The proposed hybrid representation and constraints enable robust recovery of intrinsic face properties.
- The method shows significant advantages for realistic face albedo and shape reconstruction from single images.
