A Comparison of Deep Learning Architectures for the 3D Generation Data

Abstract

There is a need to identify the best artificial images for each use case faced with several Deep Learning architectures for generating them. Twelve models with different hyperparameters were created to compare several networks with the generative architectures Autoencoder, Variational Autoencoder, and Generative Adversarial Networks in the 3D MNIST dataset. After training, the models were compared with loss functions to assess the difference between the original and artificial data, so that greater complexity did not translate into better performance, indicating the Autoencoder models as the best cost-benefit.