TSIT: A Simple and Versatile Framework for Image-to-Image Translation

TL;DR

TSIT is a straightforward, flexible image-to-image translation framework that leverages normalization layers and a two-stream model to effectively capture multi-scale semantic and style information, enabling multi-modal synthesis without extra constraints.

Contribution

Introduces a simple, versatile image translation method with a novel two-stream architecture and feature transformations, eliminating the need for cycle consistency constraints.

Findings

Effective in various tasks with high perceptual quality

Supports multi-modal style control

Outperforms several state-of-the-art baselines

Abstract

We introduce a simple and versatile framework for image-to-image translation. We unearth the importance of normalization layers, and provide a carefully designed two-stream generative model with newly proposed feature transformations in a coarse-to-fine fashion. This allows multi-scale semantic structure information and style representation to be effectively captured and fused by the network, permitting our method to scale to various tasks in both unsupervised and supervised settings. No additional constraints (e.g., cycle consistency) are needed, contributing to a very clean and simple method. Multi-modal image synthesis with arbitrary style control is made possible. A systematic study compares the proposed method with several state-of-the-art task-specific baselines, verifying its effectiveness in both perceptual quality and quantitative evaluations.