Datensatz

Zusammenfassung

The growing importance of Virtual Reality (VR) in recent years has put emphasis on the topic of avatars: Especially for shared VR experiences, it is important to provide a visual representation of the user. An important question is: What will these avatars look like? Even though VR offers great artistic freedom in the design of virtual avatars, serious applications of social VR will most likely prefer a
design close to the real appearance of humans. For this, a natural choice is the derivation of an avatar from the true appearance of a human person, using photographs or 3D scans, for example. While this will result in a close resemblance to the real person, achieving ultimate photo-realism is challenging, especially considering the typically uncontrolled recording conditions present in normal situ-
ations. Human avatars produced from such digital body captures are likely to be clearly distinguishable from photographs, which might result in negative reactions from observers due to conflicting cues. Instead of fighting the uphill battle to achieve ultimate photo-realism, we propose a deliberate deviation from realism by means of automatic stylization of such body captures. Here we recap previous
research on this topic by us and others, summarize experimental results from an automatic stylization system, and discuss potential applications and future work.