A deep learning method for generating 3D shapes with part information

What is it about?

A deep learning model trained to generate 3D shapes represented by deformable polygonal meshes. Meshes are the most popular shape representations in computer graphics. A deformable mesh can be easily edited to create more shapes. A key feature of the generated meshes is that they are already in the form of meaningful parts. Hence the deformation will be part-based which is more intuitive to modelers.

Featured Image

Why is it important?

Deep learning has achieved tremendous success in image, video, and speed analysis and synthesis. However, to train a deep learning models to generate geometry is only an emerging area. This work contributes to this effort by presenting the first generated model for deformable and structured (i.e., with parts) meshes.

Perspectives