Section: New Results

Creating and Interacting with Virtual Prototypes

• Scientist in charge: Jean-Claude Léon.

• Other permanent researchers: Marie-Paule Cani, Olivier Palombi, Damien Rohmer, Rémi Ronfard.

The challenge is to develop more effective ways to put the user in the loop during content authoring. We generally rely on sketching techniques for quickly drafting new content, and on sculpting methods (in the sense of gesture-driven, continuous distortion) for further 3D content refinement and editing. The objective is to extend these expressive modeling techniques to general content, from complex shapes and assemblies to animated content. As a complement, we are exploring the use of various 2D or 3D input devices to ease interactive 3D content creation.

Sculpting shape hierarchies

Sculpting paradigm has been successfully applied to deform simple smooth surfaces. More complex objects representing virtual characters or real-life objects are however modeled as hierarchy of shapes with elements, sub-elements and details. Applying sculpting deformation to such objects is challenging as every parts of the hierarchy should stay coherent through the deformation.

When an object can be represented as a smooth underlying surface and a set of singular details, we proposed a real-time deformation approach enabling to freely stretch or squeeze the 3D object while continuously maintaining the details' appearance. Instead of stretching or squeezing the details the same way than the smooth underlying surface, we duplicate or merge them smoothly while ensuring that the distribution of details has the same caracteristic than the original one. We published this work in Shape Modeling International [13] .

In the case of more general object hierarchies, we are developing a new methodology to apply generic deformation into complex assemblies while preserving their properties in extending the shape grammar approach into our new deformation grammar. We presented our preliminary results as a communication in the GTMG conference [35] .

Sketching and sculpting Virtual Worlds

Participants : Marie-Paule Cani, Guillaume Cordonnier, Ulysse Vimont.

Modeling virtual worlds is particular challenging: the fractal-like distribution of details in terrain shapes makes them easy to identify, but very difficult to design using standard modeling software, even for expert users. Moreover, virtual worlds involve distributions of different categories of contents over terrains, such as vegetation, houses, roads or rivers. Efficiently modeling these sets of elements, which are statistically correlated, is indeed a challenge.

This year, our contributions to tackle these issues were two-folds:

Firstly, we investigated the use of a plate tectonics metaphor to generate plausible terrains from a simple vector map representing the location of the main rivers and mountain picks. The method uses a Voronoi tesselation of pick locations to automatically generate tectonic plates which themselves drive terrain folds. Hydraulic erosion is then used to further sculpt the terrain and add details, while the specified rivers are considered to maintain consistency with the input map. This work was published in [27] . A more accurate modeling of large scale fluvial erosion and plates tectonics phenomena was investigated in Guillaume Cordonnier's master thesis and is the object of his PhD, which started in October 2015.

Secondly, we proposed a paint-based interface to tackle the problem of easily populating a terrain with distributions of objects (trees, rocks, grass, houses, etc) or of graph-like structures such as rivers and roads. The key point of our solution is to learn statistics about distributions of elements and their correlation with other distributions, with graph structures, or with terrain slope, and store the resulting histograms as "colors" in a palette interface. After creating a few local distribution manually, the user selects them with a pipette tool, and is able to reuse them with a brush. We also provided a gradient tool to interpolate between two such "colors" an move tool enabling, for instance to move groups of trees and rocks over a terrain while maintaining the adequate correlation with local slope, and a deformation interface based on seam carving enabling to seamlessly stretch or compress a region of virtual world. This work, a collaboration between Arnaud Emilien when defended his phD in December 2014, Ulysse Vimont, Marie-Paule Cani, and Bedrich Benes from Purdue University, was published at Siggraph 2015 [8] .

Sketching and sculpting Motion

Participants : Marie-Paule Cani, Martin Guay, Kevin Jordao, Rémi Ronfard.

Sketching and sculpting methods were restricted so far to the design of static shapes. One of our research goals has been to extend these interaction metaphors to motion design. This year, this included three specific contributions.

Firstly, to handle sketch-based representation of motion in the 2D case, we extended the static vector graphics complex data structure, which we had introduced at Siggraph last year, to vector graphics animations with time-varying topology [6] . This second paper was presented at Siggraph again this year. The proposed data structure is augmented with a rich set of editing operations, which can be used to quickly interpolate 2D drawings with different topologies. This work was done within a collaboration with Boris Dalstein and Michiel van de Panne from UBC, Canada.

Secondly, following a first method enabling to sculpt crowd animations (Jordao, Eurographics 2014), we developed a painting interface enabling to specify both density and main directions of motion in an animated crowd. The resulting system is still based on crowd-patches, i.e. the crowd motion is an assembly of local trajectories defined in interconnected patches. Our new painting system, called Crowd-Art, uses discrete changes in loop trajectories to evolve the number of in/out constraints in each patch until the requested density and directions are best matched. See [22] . This concluded Kevin Jordao's PhD thesis, co-advided by Julien Pettre from the MimeTIC team and in collaboration with Marc Christie, defended in December 2015.

Lastly, we developed the first expressive interface to interactively sketch and progressively sculpt and refine character motion. Our solution is based on a space-time sketching metaphor: The user sketches a single space-time stroke, which is used to initialize a series of dynamic lines of action, serving as intermediates to animate the character's model. Motion and shape deformation can be immediately replayed from this single stroke, since it sets at the same time shape, trajectory and speed (defined from the drawing speed). Thanks to visual feedback, the user can easily refine the resulting motion by editing specific lines of actions at fixed times, or by composing several motions together. This work, published at Siggraph, is one of the first methods enabling arbitrary motion to be defined from scratch by a beginner [9] . Together to another work enabling to add dynamics to character motion [21] , this concluded Martin Guay's PhD thesis, defended in June 2015.