Example-based 3D-modelling support – CrABEx

The CrABEx project is divided into five main tasks (plus a task dedicated to the project organization and management) as follow.

- Task0: Project organization and management
This task is about organization and project management.

- Task-1: Data Collection
The goal of this task is to gather and to format the database of 3D models along with their respective geometric meta-data (skeletons, cages, textures, etc.) This collection will then be enriched with as many geometric meta-data as possible.

- Task-2: Partial Shape Retrieval
Partial-shape retrieval is obviously the first tool to use when creating new content with the help of an existing database. We focus our research on progressive partial shape similarity evaluation algorithms, allowing for a continuous refinement of the similarity-based sorting of the production database.

- Task-3: Geometric Meta-Data Fitting
Once interesting examples have been identified through the shape retrieval process, their geometric meta-data (textures, animation skeleton, etc.) can be recycled onto the new input shape. But it first requires a fitting step, to adapt the meta-data to the input. Such a fitting procedure refers to the more general problem of cross parameterization.

- Task-4: Integration and User-feedback
This final task is dedicated to the integration, the evaluation and the optimization of the content-creation platform which is the core of the project, in order to produce a fully integrated and intelligent system, able to provide, at runtime, smart suggestions and a valuable assistance to the designers.

After one year and a half of work, the main results are the following.

- Task-0: According to the initial plan, the consortium agreement has been written, signer by all partners, et submitted to the ANR before the end of the 1st year of the project. The website is online and uptdated regularly. The consortium meetings are held twice a year.

- Task-1: 3D models have been gathered and the data collection is growing constantly with new models coming from public databases (SHREC databases for instance) or from private databases (3D models from the 3DDUO company ou from Gamagora). Gamagora and 3DDUO are also working on new 3D-models to be added to the collection.

- Task-2: We are working on automatic 3D-mesh segmentation and labeling methods. Such methods are the first step before working on new partial-shape retrieval algorithms. We have proposed a new semantic descriptor that is describing in a continuous manner the geometric and topological relationship of each vertex of a 3D segmented model, and labeled with each semantic label (e.g. arm, body, face for a humanoid model). We have analyzed the invariance properties of this descriptor and demonstrated its advantages within the framework of a smart assisted 3D design usage.

- Task-3: We are developing an automatic cross-parameterization method for pairs of surfaces, with a novel approach based on the joint prescription of cone singularities. We have developed an automatic method of cross-parameterization of surfaces with low distortion. This method will permit the transfer of surface meta-data (textures, normal maps, relief maps, etc.) We have developed a method derived from least square conformal maps for cross-parameterization of surfaces.

- Task-4: As originally planned in the project, the integration task has not started yet.

We foresee industrial benefits of this research at several levels, mostly in design studios related to the entertainment industry.

The main target of example-based 3D modeling support is to optimize the production pipeline of 3D content (both in term of time delay and cost). Thus, it is expectable that the productivity of the studio designers will be greatly improved, thanks to the «recycling« potential of our framework.

We also believe that such an approach can lead to new ways to envision team work in design studios. Indeed, the expertise of experienced designers will be more accessible to novice designers, who will be able to easily reproduce an expert's style and to produce high quality designs.

The envisioned system might also be seen to some extent as a new pedagogical tool for teaching 3D modeling, allowing professors to easily show and instantaneously integrate the «right things to do« on a model (based on the retrieved examples), whenever a student gets stuck in his/her design.

Vincent Léon, Nicolas Bonneel, Guillaume Lavoué, Jean-Philippe Vandeborre, Géodésiques sémantiques pour la description et la labélisation automatique de parties, 27èmes journées de l’Association Française d’Informa­tique Graphique, Reims Image 2014, Reims, France, 25-28 novembre 2014.

MINOR REVISION: Vincent Léon, Nicolas Bonneel, Guillaume Lavoué, Jean-Philippe Vandeborre, Continuous Semantic Description of 3D Meshes, CAD/Graphics 2015, August 2015, Xi'An, China, (will be published in a special section of the Elsevier journal Computers and Graphics)

The CrABEx project deals with the optimization of the production and editing chain of 3D content, in the context of digital entertainment (motion pictures, video games, serious games). Recent advances in the areas of video games, animation, physical modeling as well as in the field of 3D display and acquisition techniques lead to a significant increase of the production of 3D content (avatars, 3D characters, architectural models, etc.). The production of this massive amount of 3D content becomes a major concern for the global production chain in the design of digital entertainment products.
The challenge addressed by this project is to facilitate the creation and editing of this 3D content, with the help of a database of existing 3D models. The target of the envisioned system is to ease designers' task in the following ways: (1) With the on-line suggestion of appropriate parts during the creation/editing of a model; and (2) with the automatic generation of new content by combination of parts coming from examples of the database.
The envisioned system will be able to suggest/create not only geometrical pieces but also their associated geometric meta-data (animation skeletons, textures, deformation cages, semantic labeling, etc.). The system will rely on two important technical aspects:
- A fast and efficient approach for 3D content indexing and retrieval;
- A fast and efficient approach for the point-to-point registration of models (for the coherent assembly of existing geometry and meta-data onto the input);
Finally, the scientific results of the project will be integrated in the software used by two partners of the project: with a plugin for Maya, used by the computer video game school Gamagora (as well as in other school in the same field); and inside the video game design platform Casual Crossing Engine developed by the company 3DDUO.