Acquiring 3D Facial Data for Model Building

There exists a large gap between quality of gathered data and what 3DMM is able to synthesize.

Therefore challenges relate not to improving quality of the data but lack of publicly available data.

Then, to build a representative dataset, following questions arise

• How many faces do we need to capture in order to build a representative model?
• How can we ensure we capture natural expressions?
• Most people are not trained to perform specific expressions → Biased dataset
• What is the right sampling strategy w.r.t. age, gender, ethincity..?

Approaches to Model 3D Shape and Facial Appearance

Shape Model

Shape model defines a shape space with an associated probability distribution of parameters where

a realistic 3D face would be represented by a particular vector w in shape space

Expression Model

Expression model aims to decouple identity and expression by modeling them in separate coefficients

Appearance Model

Appearance of a face is influenced by its albedo and illumination. Appearance model capture this

information from texture in uv -space