Shape Registration

Performs shape registration on the input set of meshes.

Input

Atleast two input meshes are required as input. The template mesh onto which the rest of the meshes will be registered to, can be selected from the parameters dialog.

Output

The output meshes indicate the following Deformable registration: In this case each vertex of a target mesh(es) holds a correspondence to some vertex of the template mesh. To visualize this registration the template mesh is colored along its largest dimension and these colors are transferred to the target mesh(es) through the correspondence. For visualization purposes the colored template mesh is also provided as an output.

Rigid registration: Using the results provided by deformable registration a rigid transformation is computed in such a way that each target mesh aligns with the provided template mesh.

Thus each target output mesh contains results of both deformable registration (through vertex colors) and rigid registration (through its pose).

Description

The algorithm has only one parameter:

Number of Eigenvalues - To compute features of each mesh we use the DiffusionNet architecture. The architecture performs convolutions in the spectral domain where the value of this parameter indicates the size of the spectral basis. The default of 24 works reasonably well. Larger value could improve quality of registration although with increased compute time.