A Deep Autoencoder With Novel Adaptive Resolution Reconstruction Loss for Disentanglement of Concepts in Face Images

Among the different categories of natural images, face images are very important because of their broad range of applications. One challenging topic of face processing by computers is extracting information related to only specific concepts from face images without the help of labels. In this article, we propose a deep autoencoder model for extracting facial concepts based on their scales. A novel adaptive resolution (AR) reconstruction loss is introduced for training the autoencoder model. With the help of this new reconstruction loss, the deep autoencoder model is able to receive a real face image and compute its representation vector, which not only makes it possible to reconstruct the input image faithfully but also separates the concepts related to specific scales. We demonstrate that the autoencoder trained using the AR reconstruction loss is able to outperform benchmark models in generating faithful and high-quality reconstructions of real face images and is able to successfully transfer the facial concepts associated with a specific scale from one input image to another.