3D Convolutional Neural Network for Object Recognition

Abstract

3D Object recognition is an important task in computer vision applications. After the success of convolutional neural networks for object recognition in 2D images, many researchers have designed convolution neural network (CNN) for 3D object recognition. The state of art methods provide favourable results. However, the availability of large/dynamic 3D dataset and computational complexity of CNN are the biggest challenge in 3D CNN. In this paper, a model for object recognition problem using volumetric data representation has been proposed. The aim of this paper is to improve CNN architecture for volume based 3D objects. We implemented two separate CNN architectures and tested them on ModelNet datasets, which represent data in the form of CAD models. We compare our results with VoxNet, which is a state-of-art recognition method.

5 Figures and Tables

Cite this paper

@inproceedings{Singh20173DCN, title={3D Convolutional Neural Network for Object Recognition}, author={Rahul Singh and Ajay Mittal and Rajesh Kumar Bhatia}, year={2017} }