Eakta Jain is a PhD candidate at Carnegie Mellon University, advised by Jessica Hodgins and Yaser Sheikh. Eakta has interned at Disney Research, Pittsburgh and at the Walt Disney Animation Studios. She received her Master of Science degree from Carnegie Mellon University in 2010 and the Bachelor of Technology degree from Indian Institute of Technology Kanpur in 2006. She received the Best Paper Award at SCA 2010 for her work on augmenting hand-animation with physically based simulation. In her free time, she enjoys swimming, reading science fiction, sketching and watching animated films.

SWS, AG 4, MMCI   Info

AG Audience

English

30 Minutes

Saarbrücken

Artists who work with visual media understand that human visual attention is a limited resource and train themselves to use this resource well. In this talk, I examine how viewer attention might influence algorithms that augment visual media with digitally created effects. I will present two projects that develop this central idea.

In the first project, we leverage the techniques of 3D computer animation, such as physical simulation, for hand-drawn character animation. Researchers have attempted to compute accurate 3D reconstructions of the shape and motion of a hand-drawn character, which are meant to act as its `proxy' in the 3D environment. This task is challenging because in addition to the 2D-to-3D depth ambiguity, hand drawings also contain geometric inconsistencies. We propose that the 3D proxy for a hand-drawn character have different levels of detail, so that at each level the error terms account for quantities that might attract viewer attention. We show that the different levels of detail can be generated via three basic error terms.

In the second project, we consider the problem of retargeting comic book artwork to digital display devices by the addition of 2D camera moves on the original still pictures. We observe that artists compose their pictures to direct the viewer's attention along a deliberate route. By recording the gaze of viewers looking at the pictures, we aim to reconstruct the artist's intention in directing viewer attention. We then generate a camera move that respects the artist-intended visual route. We verify experimentally that the artist is indeed successful in directing viewer attention, and propose an algorithm to predict the parameters of a camera move on a still picture based on the recorded gaze data of multiple viewers looking at that picture.

Thorsten Thormählen

+49 681 9325-417

--email hidden

passcode not visible

logged in users only