Powers of layers for image-to-image translation

TL;DR

This paper introduces a residual block-based architecture operating in the latent space for unpaired image-to-image translation, enabling flexible, parameter-efficient transformations like style transfer and denoising.

Contribution

It presents a novel iterative residual block approach in the latent space that simplifies training and allows modulation of transformation strength, outperforming CycleGAN in some cases.

Findings

Comparable or better performance than CycleGAN

Fewer parameters needed for similar tasks

Flexible control over transformation strength

Abstract

We propose a simple architecture to address unpaired image-to-image translation tasks: style or class transfer, denoising, deblurring, deblocking, etc. We start from an image autoencoder architecture with fixed weights. For each task we learn a residual block operating in the latent space, which is iteratively called until the target domain is reached. A specific training schedule is required to alleviate the exponentiation effect of the iterations. At test time, it offers several advantages: the number of weight parameters is limited and the compositional design allows one to modulate the strength of the transformation with the number of iterations. This is useful, for instance, when the type or amount of noise to suppress is not known in advance. Experimentally, we provide proofs of concepts showing the interest of our method for many transformations. The performance of our model is comparable or better than CycleGAN with significantly fewer parameters.