A Comprehensive Study of Autoencoders` Applications Related to Images

Abstract

This article incorporates a comprehensive study of autoencoders` applications related to images. First of all, a vanilla autoencoder is described along with details of its architecture and training procedure. Secondly, main methods for regularization of it are exposed, such as dropout and additive gaussian noise. The applications of autoencoders such as image morphing, reconstruction and search are shown. Then, the VAE (variational autoencoder) is highlighted. Main applications of it such as outliers detection and image generation are described. Finally, it`s shown that using warm-up for VAE with respect to KL loss gives much more plausible results in terms of image generation.