No document available.
Abstract :
[en] The DIVA-Face database is a collection of 8,064 video clips and 2,016 photographs of eight Caucasian actresses.
Each video clip depicts a single actress performing one out of twelve standardized gestures under one out of three lighting conditions (front-, left-, or right-illumination), within one out of twenty-eight appearance configurations. The twenty-eight appearance configurations were systematically created by filming each actress with one out of six wigs (specific to each pair) or wearing a balaclava, with and without glasses, and with and without makeup (lipstick). For added variability, four wigs per pair were filmed in two hairstyles (loose or tied up). Each appearance was recorded performing all twelve gestures in the three lighting conditions.
In addition, actresses were photographed in each appearance–lighting combination, with a neutral facial expression, and while facing in three distinct directions (front, three-quarter left, and three-quarter right).
Therefore, the collection of stimuli allows a controlled selection of facial stimuli based on each characteristic (wig/balaclava, presence of make-up, presence of glasses, lighting, gesture or viewing angle) independently.
The actresses are grouped into four pairs, matched for average facial distinctiveness and pairwise facial similarity based on ratings from independent judges. Within each group, actresses were filmed and photographed with the same subset of six wigs out of 24. Four distinct subset of wigs were created based on distinctiveness and similarity ratings.
All stimuli were recorded against a green screen, allowing easy manipulation or replacement of the background. The database is designed for flexible use in psychophysics, neurocognitive, or socio-cognitive experiments involving facial stimuli.
