Table of Contents
Fetching ...

Eyelid Fold Consistency in Facial Modeling

Lohit Petikam, Charlie Hewitt, Fatemeh Saleh, Tadas Baltrušaitis

TL;DR

This work proposes a new definition of eyelid fold consistency and implements geometric processing techniques to model diverse eyelid shapes in a unified topology and reprocess data used to train a parametric face model and demonstrates significant improvements in face-related machine learning tasks.

Abstract

Eyelid shape is integral to identity and likeness in human facial modeling. Human eyelids are diverse in appearance with varied skin fold and epicanthal fold morphology between individuals. Existing parametric face models express eyelid shape variation to an extent, but do not preserve sufficient likeness across a diverse range of individuals. We propose a new definition of eyelid fold consistency and implement geometric processing techniques to model diverse eyelid shapes in a unified topology. Using this method we reprocess data used to train a parametric face model and demonstrate significant improvements in face-related machine learning tasks.

Eyelid Fold Consistency in Facial Modeling

TL;DR

This work proposes a new definition of eyelid fold consistency and implements geometric processing techniques to model diverse eyelid shapes in a unified topology and reprocess data used to train a parametric face model and demonstrates significant improvements in face-related machine learning tasks.

Abstract

Eyelid shape is integral to identity and likeness in human facial modeling. Human eyelids are diverse in appearance with varied skin fold and epicanthal fold morphology between individuals. Existing parametric face models express eyelid shape variation to an extent, but do not preserve sufficient likeness across a diverse range of individuals. We propose a new definition of eyelid fold consistency and implement geometric processing techniques to model diverse eyelid shapes in a unified topology. Using this method we reprocess data used to train a parametric face model and demonstrate significant improvements in face-related machine learning tasks.

Paper Structure

This paper contains 14 sections, 15 figures.

Figures (15)

  • Figure 1: For an individual with a hooded eyelid and epicanthal fold; FLAME FLAME:SiggraphAsia2017 cannot model either in geometry. hewitt2024look do not model hoodedness causing an over-smoothed eyelid crease. Our definition ensures eyelid creases are sharp and that hoodedness and epicanthal folds are explicitly modeled in geometry. Orange depicts the surface beneath the eyelid crease. Photo $\copyright$ Triplegangers
  • Figure 2: Topological inconsistencies in eyelids of training data of hewitt2024look. The same geometric feature (yellow) is not always represented by the same edge-loop (blue).
  • Figure 3: Left: Hooded and monolid eyelids without a prominent eyelid fold. Middle: Partially closed revealing the fold. Right: Our annotation acknowledging that the fold is just behind the eyelid hood. Videos $\copyright$ AILA_IMAGES, Damato / Adobe Stock
  • Figure 4: Eyelid shapes (top) with annotations (bottom) based on our definition of eyelid consistency.
  • Figure 5: Template meshes in the topology of hewitt2024look with explicitly modeled eyelid creases on shared edge-loop (cyan). The far-right template has an explicitly modeled crease just behind the eyelid hood and epicanthal fold (pink).
  • ...and 10 more figures