A Hybrid Deep Learning Network CNN-SVM for 3D Mesh Segmentation

3D shape segmentation is considered to be one of the critical tasks in computer vision and graphics. With the wider availability of mesh data, deep learning has established itself as a powerful technique in 3D mesh segmentation and classification by demonstrating excellent performances. In this paper, we implement a deep architecture to segment and label 3D shape parts. The proposed pipeline is based on a Convolutional Neural Network (CNN) connected to a Linear Support Vector Machine (LSVM) that is simple and useful for the classification tasks. To learn objective models, we feed the designed pipeline with a combination of geometric features and spatial information based on Classical representative features. Experiments are performed using "3D mesh segmentation benchmark" database to show the performance of the proposed approach.