TSIT: A Simple and Versatile Framework for Image-to-Image Translation
This work addresses the problem of versatile and simple image synthesis for computer vision applications, offering a clean method that scales to both unsupervised and supervised settings.
The paper tackles image-to-image translation by proposing a two-stream generative model with novel feature transformations, achieving competitive perceptual quality and quantitative results across various tasks without needing additional constraints like cycle consistency.
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.