Example-Based Facial Rigging

Sentence: Method for generating optimal facial blendshape rigs from a set of example poses transferred from a generic template to a target blendshape model.
Problem: In order to get realistic facial animation, artists typically need to sculpt hundreds of individual blend shapes, which is an enormously time-consuming process. It would be much easier to start with blendshapes for all the ‘standard’ expression and then build/customize from there (don’t reinvent the wheel).
Key: The expression is transferred from the template model by mapping the deformation gradients (between the triangles of the template’s neutral expression and the triangles of the new expression) onto the target model.
Basic Info: In order to transfer an expression from a template to an arbitrary target model, two problems need to be solved.
1) How to generate target blendshapes that match the geometry and motion of the actor given on a few training poses. This is done by mapping the deformation gradients (as mentioned above).
2) How to achieve blending weights so expressions are semantically similar for the template and the target. This is done by using least-square regression on the generated blenshapes until the ideal deformation is found.
Contribution: greatly increase the productivity of professional artists with quick generation of actor-specific blendshapes. Also, creates an opportunity for non-professionals (without time or expertise) to easily make facial animations.
Resources: Video & PDF