Face Interpretation with deep and ensemble Learning – FacIL

Facial expressions convey information regarding a subject's cognitive state as well as its social intentions. As such, facial expression recognition is an active domain on the edge between computer vision and machine learning that aims at automatically extracting such cues from an image or a video stream. In the last fifteen years, research in facial expression recognition has been through significant breakthroughs. Today, the technology is mature enough to open the door for commercial application in several domains such as markerless facial movement capture, automatic analysis of emotional behavior for marketing or facial expression monitoring for behavioral studies.
However, to be truly effective and spread on a larger scale, the next generation of automatic facial expression recognition systems need to overcome some major challenges: to deal with the great variability of real-world data and the lack labeled data, and to design flexible models with low computational complexity. Models will be validated through two real-world use cases: the quality assessment of facial emotion produced by children playing to a serious game and the characterization of respiratory distress (dyspnea) of patients under respiratory assistance. Moreover an open source software library that gathers the models developed in the frame of the project will be released to the community