Valentin Vielzeuf




Applied automatic emotion recognition from facial expression

Automatic emotion recognition from facial expression is currently a broad topic of interest, finding several applications such as consumer behavior analysis, psychiatric and especially autism research or even animation movies tools. Coming from these different fields, there are several ways to represent a facial expression, from simple words ("angry", "sad") to more complex notions (arousal/valence, action units).

First, the presentation will review these different representations, detailing their pros and cons.
Then we will present a deep learning framework which allows to gather all of them into a more compact and generic representation.

The benefits of the so obtained representation will finally be evaluated in two use cases :
- Facial expression recognition across several datasets
- Transfer learning for pain estimation