Gesture3D: Posing 3D Characters via Gesture Drawings

摘要

Artists routinely use gesture drawings to communicate ideatedrncharacter poses for storyboarding and other digital media. Duringrnsubsequent posing of the 3D character models, they use theserndrawing as a reference, and perform the posing itself using 3D interfacesrnwhich require time and expert 3D knowledge to operate.rnWe propose the first method for automatically posing 3D charactersrndirectly using gesture drawings as an input, sidestepping the manualrn3D posing step. We observe that artists are skilled at quicklyrnand effectively conveying poses using such drawings, and designrnthem to facilitate a single perceptually consistent pose interpretationrnby viewers. Our algorithm leverages perceptual cues to parsernthe drawings and recover the artist-intended poses. It takes as inputrna vector-format rough gesture drawing and a rigged 3D characterrnmodel, and plausibly poses the character to conform to the depictedrnpose. No other input is required. Our contribution is two-fold:rnwe first analyze and formulate the pose cues encoded in gesturerndrawings; we then employ these cues to compute a plausible imagernspace projection of the conveyed pose and to imbue it with depth.rnOur framework is designed to robustly overcome errors and inaccuraciesrnfrequent in typical gesture drawings. We exhibit a wide varietyrnof character models posed by our method created from gesturerndrawings of complex poses, including poses with occlusions andrnforeshortening. We validate our approach via result comparisons tornartist-posed models generated from the same reference drawings,rnvia studies that confirm that our results agree with viewer perception,rnand via comparison to algorithmic alternatives.