Section: New Results

Creating and Interacting with Virtual Prototypes

• Scientist in charge: Jean-Claude Léon.

• Other permanent researchers: Marie-Paule Cani, Frédéric Devernay, 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 Virtual Worlds

Extending expressive modeling paradigms to full virtual worlds, with complex terrains, streams and oceans, and vegetation is a challenging goal. To achieve this, we need to combine procedural methods that accurately simulate physical, geological and biological phenomena shaping the world, which high level user control. This year, our work in the area was three-folds:

Firstly, in collaboration with Jean Braun, professor in geo-morphology and other colleagues, we designed the fist efficient simulation method able to take into account large-scale fluvial erosion to shape mountains. This method was published at Eurographics [10]. We also designed an interactive sculpting system with multi-touch finger interaction, able to shape mountain ranges based on tectonic forces. This method, combined in real-time with our erosion simulation process, was submitted for a journal publication.

Secondly, we extended the "Worldbrush" system proposed in 2015 (Emilien et al, Siggraph 2015) in order to consistently populate virtual worlds with learned statistical distributions of trees and plants. The main contributor to this project was James Gain, our visiting professor. After clustering the input terrain into a number of characteristic environmental conditions, we computed sand-box (small-scale) simulations of ecosystems (plant growth) for each of these conditions, and then used learned statistical models (an extension of worldbrush) to populate the full terrain with consistent sets of species. This work was submitted for publication.

Third, we extended interactive sculpting paradigms to the sculpting of liquid simulation results, such as editing waves on a virtual ocean [23]. Liquid simulations are both compute intensive and very hard to control, since they are typically edited by re-launching the simulations with slightly different initial conditions until the user is satisfied. In contrast, our method enables users to directly edit liquid animation results (coming in the form of animated meshes) in order to directly output new animations. More precisely, the method offer semi-automatic clustering methods enabling users to select features such as droplets and waves, edit them in space and time and them paste them back into the current liquid animation or to another one.

Sketch based design

Using 2D sketches is one of the easiest way for creating 3D contents. While prior knowledge on the object being sketch can be used to retrieve the missing information, and thus consistently inferring 3D, interpreting more general sketches and generating 3D shapes from them in indeed a challenging long-term goals. This year, our work in the area was two-folds:

Firstly, we participated to a course on Sketch-based Modeling, presented at both Eurographics 2016 and Siggraph Asia 2016 [18]. The parts we worked on was sketch-based modeling from prior knowledge, with the examples of our works on animals, garments (developable surfaces) and trees.

Secondly, we advanced towards the interpretation of general sketches representing smooth, organic shapes. The key features of our methods are a new approach for aesthetic contour completion, and an interactive algorithm for progressively interpreting internal silhouettes (suggestive contours) in order to progressively extract sub-parts of the shape from the drawing. These parts are ordered in depth. Our first results were presented as a poster at the Siggraph 2016 conference [37], and then extended an submitted for publication. We are now extending them towards the inference of 3D, organic shapes from a 2D sketch.