# Latent Filter Scaling for Multimodal Unsupervised Image-to-Image Translation

## TL;DR

This paper introduces a simple, hyperparameter-free method for multimodal unsupervised image-to-image translation that improves image quality and disentangles content and style, outperforming current state-of-the-art approaches.

## Contribution

The proposed method treats the latent code as a filter modifier, simplifying architecture and hyperparameter tuning while enhancing image quality and enabling disentanglement.

## Key findings

- Outperforms state-of-the-art in image quality
- Maintains multimodal diversity without extra hyperparameters
- Achieves disentanglement of content and style

## Abstract

In multimodal unsupervised image-to-image translation tasks, the goal is to translate an image from the source domain to many images in the target domain. We present a simple method that produces higher quality images than current state-of-the-art while maintaining the same amount of multimodal diversity. Previous methods follow the unconditional approach of trying to map the latent code directly to a full-size image. This leads to complicated network architectures with several introduced hyperparameters to tune. By treating the latent code as a modifier of the convolutional filters, we produce multimodal output while maintaining the traditional Generative Adversarial Network (GAN) loss and without additional hyperparameters. The only tuning required by our method controls the tradeoff between variability and quality of generated images. Furthermore, we achieve disentanglement between source domain content and target domain style for free as a by-product of our formulation. We perform qualitative and quantitative experiments showing the advantages of our method compared with the state-of-the art on multiple benchmark image-to-image translation datasets.