Abstract :
[en] Understanding how humans learn, recognize or otherwise process unfamiliar faces requires stimulus materials that balance experimental control with ecological validity. Although face learning research has highlighted the importance of within-person variability, most existing studies rely on static ambient images or do not systematically disentangle appearance-based variability from other display-related factors. Here, we introduce the Dynamic and Image Variation Archive (DIVA-Face), a video-based face database designed to fill this gap. The database comprises 8,064 video clips and 2,016 still images of eight actresses recorded under tightly controlled conditions. Each identity is presented across orthogonally manipulated appearance dimensions, including hairstyle (via six distinct wigs—hair down or up—and one balaclava), makeup (lipstick present or absent), and accessory (glasses present or absent), giving 28 distinct combinations. Each appearance combination was captured under three lighting directions, while performing twelve dynamic facial actions for videos, or posing with three head orientations for photographs. All recordings were made against a uniform green-screen background, allowing flexible manipulation of contextual information for future applications. To validate the usability of the database, we conducted a preregistered proof-of-concept experiment using an incidental face learning paradigm with a subset of the stimuli. Results show reliable effects of appearance manipulations on recognition performance, even under online testing conditions. These findings confirm that the database supports controlled behavioral investigations of unfamiliar face learning. The database is also suited for behavioral, social-cognitive, and computational investigations beyond face learning, and is openly available to promote reproducibility and cross-disciplinary integration.
