Catboost3D: A Novel CatBoost-based Approach for Efficient Classification of 3D Models

Abstract

The classification of 3D models consists in grouping similar objects in predefined classes. This task is challenging and requires results that align with human perception. This paper introduces a new classification method for 3D models, using the CatBoost classifier, a machine learning algorithm that has shown good results in various domains. Our approach addresses the challenge of accurately categorizing 3D models without needing a large database for training and using the 3D models directly without passing by intermediate representation. We trained the CatBoost classifier using the dihedral angle, shape index, and shape diameter function to create a model able to classify each query object to the correct class. We demonstrate, through extensive experimentation, that our method not only accurately classifies 3D models into their respective categories but also effectively handles the complexity and variability inherent in 3D data.