Projet MADRAS / MADRAS Project

3D Models And Dynamic models Representation And Segmentation


PhD Thesis – subject #2

Perception-based segmentation of dynamic meshes for compression and transmission.


Main contact: Franck Hétroy
  Franck dot Hetroy at imag dot fr (anti-spam: please correct the e-mail address)
  direct phone number: +33 3 (0)476 615 504

Practical information

The PhD student will work within the MADRAS project. The grant will be about 1500 euros per month during three years.

The PhD student will start his thesis at INRIA Grenoble, where the EVASION team is located. He will also have to spend some time in the M2DisCo team of the LIRIS lab in Lyon.


For a few years, sequences of 3D meshes varying through time, which are often called dynamic meshes, have become more and more popular in the industry, e.g. in the entertainment area. These dynamic meshes need to be analyzed and stored for various applications, the first of them being compression and transmission since the amount of data quickly becomes huge (a 10 second-sequence with a frame rate of 30 images per second contains 300 meshes !).

In the MPEG-4 standard, the coding of dynamic meshes uses 3D mesh coding (3DMC) for the first mesh and Interpolator Compression (AFX-IC) for the animation part. The few works that deal with compression or transmission of sequences of 3D meshes mainly focus on multiresolution analysis [1,2]. However, these approaches often have strong requirements about the mesh sequence: for instance, each mesh should be semi-regular or have a given connectivity.


This thesis explores another approach. As explained by Lengyel as early as 1999 [3], if the first mesh of the sequence (or a set of "base" meshes) is segmented into smaller submeshes whose motion can be easily described, for instance as rigid-body, it is necessary only to specify the affine transform for each part instead of the motion of each vertex. For example when we deal with character animation, the possible motion is restricted and the segmentation should approximately correspond to the character anatomical structure: for instance, all vertices related to a given forearm should move the same way, since this arm can be considered as rigid.

The goal of this thesis is to propose new ways to segment dynamic meshes without temporal coherency (that is to say, there is no bijection between vertices of the mesh at time t and vertices of the mesh at time t+dt), which allows efficient compression and transmission. The proposed segmentation method(s) will use visual perception criteria to enhance compression.

The main objectives of this PhD thesis are: