Directional GAN: A Novel Conditioning Strategy for Generative Networks

TL;DR

Directional GAN introduces a new conditioning method that modifies latent vectors with semantic attribute directions, enabling controlled image generation without retraining the generator, applicable to various attribute types.

Contribution

The paper presents a simple, effective conditioning strategy for GANs that leverages directional vectors in latent space, enhancing controllability without retraining.

Findings

Achieved an average attribute accuracy of 86.4% across datasets.

Applicable to both discrete and continuous attributes.

Works with pre-trained unconditional GANs.

Abstract

Image content is a predominant factor in marketing campaigns, websites and banners. Today, marketers and designers spend considerable time and money in generating such professional quality content. We take a step towards simplifying this process using Generative Adversarial Networks (GANs). We propose a simple and novel conditioning strategy which allows generation of images conditioned on given semantic attributes using a generator trained for an unconditional image generation task. Our approach is based on modifying latent vectors, using directional vectors of relevant semantic attributes in latent space. Our method is designed to work with both discrete (binary and multi-class) and continuous image attributes. We show the applicability of our proposed approach, named Directional GAN, on multiple public datasets, with an average accuracy of 86.4% across different attributes.