DINO: A Conditional Energy-Based GAN for Domain Translation

TL;DR

This paper introduces DINO, a novel framework for domain translation using a pair of networks trained simultaneously, which better preserves shared semantics and is effective across various challenging cross-modal translation tasks.

Contribution

The paper presents a new dual-network framework for domain translation that improves semantic preservation and broad applicability over traditional conditional GAN approaches.

Findings

Outperforms existing methods in preserving shared semantics.

Effective in challenging cross-modal translation tasks.

More generic and adaptable to various problems.

Abstract

Domain translation is the process of transforming data from one domain to another while preserving the common semantics. Some of the most popular domain translation systems are based on conditional generative adversarial networks, which use source domain data to drive the generator and as an input to the discriminator. However, this approach does not enforce the preservation of shared semantics since the conditional input can often be ignored by the discriminator. We propose an alternative method for conditioning and present a new framework, where two networks are simultaneously trained, in a supervised manner, to perform domain translation in opposite directions. Our method is not only better at capturing the shared information between two domains but is more generic and can be applied to a broader range of problems. The proposed framework performs well even in challenging cross-modal translations, such as video-driven speech reconstruction, for which other systems struggle to maintain correspondence.