BlendFields: Few-Shot Example-Driven Facial Modeling

  • 2023-05-12 15:30:07
  • Kacper Kania, Stephan J. Garbin, Andrea Tagliasacchi, Virginia Estellers, Kwang Moo Yi, Julien Valentin, Tomasz TrzciƄski, Marek Kowalski
  • 21


Generating faithful visualizations of human faces requires capturing bothcoarse and fine-level details of the face geometry and appearance. Existingmethods are either data-driven, requiring an extensive corpus of data notpublicly accessible to the research community, or fail to capture fine detailsbecause they rely on geometric face models that cannot represent fine-graineddetails in texture with a mesh discretization and linear deformation designedto model only a coarse face geometry. We introduce a method that bridges thisgap by drawing inspiration from traditional computer graphics techniques.Unseen expressions are modeled by blending appearance from a sparse set ofextreme poses. This blending is performed by measuring local volumetric changesin those expressions and locally reproducing their appearance whenever asimilar expression is performed at test time. We show that our methodgeneralizes to unseen expressions, adding fine-grained effects on top of smoothvolumetric deformations of a face, and demonstrate how it generalizes beyondfaces.


Quick Read (beta)

loading the full paper ...