SMI'10-Shape Modeling International 2010, Arts et Métiers ParisTech, LSIS-Laboratory of Information and Systems Sciences

Professor Marie Paule Cani


Professor of Computer Science at the Grenoble Institute of Technology, France Marie-Paule Cani is the head of the INRIA research group EVASION, a team of Laboratoire Jean Kuntzmann (CNRS, Grenoble University).

Her main research interest is the creation of digital content for animated virtual worlds, which leads to developing techniques for both shape design and animation. Her contributions include algorithms for efficient physically-based animation,the use of implicit surfaces and of constant volume space deformations in interactive sculpting systems, 3D clothing design from sketches, physically-based models such as super-helices for hair animation and the design of layered models for efficiently animating natural scenes.

Marie-Paule Cani served as conference or programm co-chair of many international conferences such as EUROGRAPHICS 2004, SMI 2005, SCA 2006, SBIM 2008 and has been associate editors of several journals (Graphical Models, IEEE TVCG, CGF). Member of the Institut Universitaire de France and Eurographics fellow, she received the national Irène Joliot Curie award for women in Computer Science in 2007.

Program : “Implicit surfaces: Adapted blends for shape modelling and animation”

Smooth blending is one of the main features of functionally-based implicit surfaces, enabling constructive modeling systems where components progressively blende and the animation of fluid-like objects that merge and splits over time. However, their practical suffers from three major issues:

- Firstly, implicit surfaces blend at distance, which makes shapes difficult to control and generates surprizing behaviours in animation;

- Secondly, the extent of the blend is difficult to control: modelling an arm which blends with the body only at one extremity is intricate;

- Thirdly, small, sharp objects tend to blur when blended into larger ones, leading to only simple, blobby shapes.

The first part of this talk presents novel solutions to these three long lasting problems. The key idea is to automatically restrict blends to the regions where implicit models intersect, while other parts of the shapes can come close to each other without being deformed.

In a second part, we discuss the use of implicit surfaces for modelling free form shapes. We first show the ability of skeleton-based, convolution surfaces to fit arbitrary 2D regions. This leads to optimization-free methods for reconstructing organic shapes from only a few cross-sections and also enables progressive sketch-based modelling of complex shapes.

Lastly we illustrate the use of implicit surfaces in animation. We show that the new blending operator can be tuned to generates plausible, progressive blends when two soft objects come in contact. In contrast, we exploit the ability of implicit primitives to blend at distance to enhance cloth animations with realistic folds.