Show, Attend and Translate: Unsupervised Image Translation with Self-Regularization and Attention

TL;DR

This paper introduces a simple yet effective unsupervised image translation model that uses self-regularization and attention mechanisms to produce high-quality translations while avoiding artifacts.

Contribution

It proposes a novel image translation approach combining self-regularization and attention modules, improving translation quality and enabling applications like unsupervised segmentation.

Findings

Outperforms existing methods in quality and simplicity

Attends to key image regions to prevent artifacts

Enables unsupervised segmentation and saliency detection

Abstract

Image translation between two domains is a class of problems aiming to learn mapping from an input image in the source domain to an output image in the target domain. It has been applied to numerous domains, such as data augmentation, domain adaptation, and unsupervised training. When paired training data is not accessible, image translation becomes an ill-posed problem. We constrain the problem with the assumption that the translated image needs to be perceptually similar to the original image and also appears to be drawn from the new domain, and propose a simple yet effective image translation model consisting of a single generator trained with a self-regularization term and an adversarial term. We further notice that existing image translation techniques are agnostic to the subjects of interest and often introduce unwanted changes or artifacts to the input. Thus we propose to add an attention module to predict an attention map to guide the image translation process. The module learns to attend to key parts of the image while keeping everything else unaltered, essentially avoiding undesired artifacts or changes. The predicted attention map also opens door to applications such as unsupervised segmentation and saliency detection. Extensive experiments and evaluations show that our model while being simpler, achieves significantly better performance than existing image translation methods.