Generative and Discriminative Voxel Modeling with Convolutional Neural Networks
This work addresses the challenge of efficient 3D data representation for applications like shape modeling and object classification, offering incremental improvements in classification performance.
The paper tackled the problem of representing three-dimensional data by exploring voxel-based models for shape modeling and object classification, achieving a 51.5% relative improvement in state-of-the-art object classification on the ModelNet benchmark.
When working with three-dimensional data, choice of representation is key. We explore voxel-based models, and present evidence for the viability of voxellated representations in applications including shape modeling and object classification. Our key contributions are methods for training voxel-based variational autoencoders, a user interface for exploring the latent space learned by the autoencoder, and a deep convolutional neural network architecture for object classification. We address challenges unique to voxel-based representations, and empirically evaluate our models on the ModelNet benchmark, where we demonstrate a 51.5% relative improvement in the state of the art for object classification.