Hide/Show Apps

Supervised mesh segmentation for 3D objects with graph Convolutional neural networks

Perek, Emir Kaan
Mesh segmentation is a fundamental application that is primarily used for understanding and analyzing 3D shapes in a broad range of areas in Computer Science. With the increasing trend of deep learning, there have been many learning-based solutions to the mesh segmentation problem based on the classification of the individual mesh polygons. In this thesis, we cast mesh segmentation as a supervised graph labeling problem by using Graph Convolutional Neural Networks (GCNN). We treat a mesh object as a graph to be labeled and develop a segmentation model that takes an entire structure of the object as input and returns its segmentation. While similar models focus on labeling each polygon of the 3D objects separately, our model is capable of labeling all polygons in a single run thanks to GCNN. Moreover, being able to use connectivity information in the graph provides an opportunity for a drastic decrease in the required features used as input to the model, compared to the previous studies. We train and test our model for segmenting human shapes, one of the challenging shapes to segment. We report competitive results compared to other state-of-art supervised segmentation techniques by using noticeably less input features.