Surface reconstruction from point clouds

In this project, we will take a closer look into surface reconstruction from point clouds. This project is a hands-on project in PyTorch. You will gain familiarity with 3D surface reconstruction and applying deep neural networks to meshes. The project is mainly focused on the Point2Mesh [1] technique, which optimizes the weights of a CNN to deform some initial mesh to shrink-wrap the input point cloud. You will apply this technique to scans and a wider variety of data in order to explore the current strengths and limitations of such a technique, and how it can be further improved. One interesting direction is to learn which edge to split, thereby defining where to add additional mesh connectivity. By the end of the project, you should better understand these topics from a theoretical and practical perspective, and gain insights with respect to how to incorporate similar concepts into your own research.

[1] Point2Mesh: A Self-Prior for Deformable Meshes. Rana Hanocka, Gal Metzer, Raja Giryes, and Daniel Cohen-Or. SIGGRAPH 2020.

Project timezone: to be decided