Colloquium DIRO -- Exploring the complementarity of computer graphics and computer vision

12/03/2015 - 15:30
12/03/2015 - 16:30
Jean-Charles Bazin (Disney Research, Zürich)
Université de Montréal, Pavillon André-Aisenstadt, 2920 ch. de la Tour, Salle 3195

Computer graphics (CG) and computer vision (CV) nicely complement each other. I am convinced that exploring their complementarity space provides exciting research opportunities and opens a wide range of novel applications. In this talk, I will present some of our projects exploring this complementarity, and hope to convince you about the great potential of combining techniques and concepts of CV and CG.

First, I will discuss about our works on 3D point processing. I will start by presenting our approach on RGBD data acquired by multiple color+depth sensors in the context of free-viewpoint video generation (Eurographics 2014). We used CV techniques (such as camera calibration and image processing) and combined them with point-based surface representation from the CG community that we extended with visual information (point color and image structure). Then I will show one of our exciting follow-up projects on scene-space video processing (SIGGRAPH 2015). Our approach tackles noisy 3D point information by leveraging the high amount of redundant visual data in videos for numerous applications such as video denoising, deblurring, and inpainting, but also computational shutter functions and action shot effects.Then I will show novel original applications also enabled by combining CV and CG concepts. First, I will describe a method for object motion manipulation in real footages (under submission) that associates physics-based simulation from the CG community and object pose estimation, camera calibration and visual constraints from CV. Other examples include music retargeting for video temporal resizing (Eurographics 2013), as well as synchronization and blending of facial performance videos (ICCV 2015) for non-linear video story-telling and interactive face editing.Finally I will present video manipulation applications at the boundary of CV and CG: real-time eye contact correction in video calls by depth and facial template manipulation (SIGGRAPH Asia 2012), and video stabilization for virtual visit in the context of cultural preservation (MVA 2011, CVPR 2012).The presentation will be given in French.

Last edited by on Wed, 12/02/2015 - 10:49