SDIT: Scalable and Diverse Cross-domain Image Translation

TL;DR

SDIT introduces a unified generator for scalable and diverse image-to-image translation, leveraging domain conditioning, latent variables for diversity, and attention mechanisms to improve focus on domain-specific features, demonstrated on face and other datasets.

Contribution

The paper presents SDIT, the first method to combine scalability and diversity in a single image translation model using domain conditioning and attention mechanisms.

Findings

Effective on face mapping datasets.

Outperforms existing methods in diversity and scalability.

Works well on non-face datasets.

Abstract

Recently, image-to-image translation research has witnessed remarkable progress. Although current approaches successfully generate diverse outputs or perform scalable image transfer, these properties have not been combined into a single method. To address this limitation, we propose SDIT: Scalable and Diverse image-to-image translation. These properties are combined into a single generator. The diversity is determined by a latent variable which is randomly sampled from a normal distribution. The scalability is obtained by conditioning the network on the domain attributes. Additionally, we also exploit an attention mechanism that permits the generator to focus on the domain-specific attribute. We empirically demonstrate the performance of the proposed method on face mapping and other datasets beyond faces.