MSR Vision Faculty Summit - Artistic Stylization of Consumer Media Collections

Опубликовано 12 августа 2016, 0:01

Recent years have seen an explosive growth in personal digital media collections, yet post-capture this content frequently lies dormant on usersΓÇÖ PCs. We present a system to breathe life into home digital media collections, drawing upon artistic stylization to create a ΓÇ£Digital Ambient DisplayΓÇ¥ that automatically selects, stylizes and transitions between digital content in a semantically meaningful sequence. To do so, we must interpret the visual structure in images and video. Specific methods covered in this talk include a new probabilistic video segmentation algorithm based on motion diffused priors propagated forward from previous frames, incorporated into a CRF framework [1]. We outline how these temporally coherent region maps can be used to both stylize video into cartoons and paintings, and measure the visual similarity between frames to enable transitions between media items [2]. The talk will focus primarily on video segmentation and stylization, however our approach to sequencing the display of visual media items will also be briefly outlined, and a preview given of our most recent results in facial portrait stylization.