Section: New Results

Knowledge-based Models for Narrative Design

  • Scientist in charge: Rémi Ronfard.

  • Other permanent researchers: Marie-Paule Cani, François Faure, Jean-Claude Léon, Olivier Palombi.

Our long term goal is to develop high-level models helping users to express and convey their own narrative content (from fiction stories to more practical educational or demonstrative scenarios). Before being able to specify the narration, a first step is to define models able to express some a priori knowledge on the background scene and on the object(s) or character(s) of interest. Our first goal is to develop 3D ontologies able to express such knowledge. The second goal is to define a representation for narration, to be used in future storyboarding frameworks and virtual direction tools. Our last goal is to develop high-level models for virtual cinematography such as rule-based cameras able to automatically follow the ongoing action and semi-automatic editing tools enabling to easily convey the narration via a movie.

Knowledge representation through 3D ontologies

Participants : Armelle Bauer, Jean-Claude Léon, Olivier Palombi.

Figure 9. Illustration from [22] showing our results on ontology based anatomy generation.

We chose to develop 3D ontologies for being able to express combined knowledge on geometry, motion and function for assemblies or hierarchies of 3D objects. This is done in collaboration with a specialized group from the LIG laboratory in Grenoble. We decided to first focus these ontologies developments on two topics on which group members have a strong expertise: the anatomical domain (an interesting application test-bed for educational scenarios) and the industrial prototyping domain (where assembly scenarios can be defined).

We developed an anatomical knowledge database called My Corporis Fabrica (MyCF). We first linked functional entities defined in MyCF to the involved anatomical structures, using the musculoskeletal system as a test-bed. Based on this new formal description of the functional anatomy of limbs, we presented a novel pipeline for the construction of biomechanical simulations by combining generic anatomical knowledge with specific data which can handle complex reasoning and querying in MyCF. This resulted into a publication in the Journal of Biomedical Semantics [11] . We also used MyCF within our previous framework of anatomical transfert to set up an assistant tool for modeling and simulating anatomical structure such as bones, muscles, viscera and fat tissues easily while ensuring a correct anatomical consistency [22] .

Secondly, in analysing the similarities and differences between existing ontology based describtion of products and virtual humans, we developed a common framework for combining 3D models and functional describtion to both models [15] , [34] .

Virtual direction tools

Participants : Adela Barbulescu, Rémi Ronfard.

Figure 10. Illustration from [20] showing our results on expressive virtual actors.

We are developing a new approach to transfer speech signals and 3D facial expressions to virtual actors of a different identity. The converted sequences should be perceived as belonging to the target actors. This is the goal of Adela Barbulescu's thesis, co-advised by Gérard Bailly from GIPSA-lab. Our work started with conversion of speaking styles through speech signals only. This year, we started extending this approach to visual prosody and advanced on communicating social attitudes through head gestures [20] .

Virtual cinematography

Participants : Quentin Galvane, Vineet Ghandi, Christophe Lino, Rémi Ronfard.

Figure 11. Illustration from [23] showing our results on narrative-driven camera control of computer games.

Our goal is to model automatic cameras for covering 3D scenes, as well as to develop semi-automatic film editing techniques to help conveying narration. This work was first conducted on video data, enabling us to test our ideas without the need for complex 3D movies: we designed an automatic method for the identification of actors in a video, and are using it for the automatic re-framing and editing high-resolution videos shots of theater rehearsals [25] .

In parallel, we started extending this methodology to 3D animation, in collaboration with the Mimetic group in Rennes and with Geneva University: this year, we proposed a new method for replaying first person video games with automatic camera control based on the narration [23] . We also advanced towards semi-automatic film editing: A paper was just accepted to AAI 2015. To stress the difficulty of validating film editing methods, we devoted a specific work to validation methodologies [27] .

We also addressed other issues related to cinematography and narratives: We designed a pre-visualization system for 3D cinematography to be used in the Action3DS project [30] : the method makes use of 3D modeling to show what the spectators watching a 3D movie are going to see, in order to ease 3D camera control by the film director. Lastly, we worked on computer generation of narrative discourses with the university of Geneva [31] .

This year, Remi Ronfard and Vineet Gandhi wrote a patent application "Dispositif de génération de rushes cinématographiques par traitement vidéo", demande de brevet français no. 1460957, déposée le 13 novembre 2014.