High Fidelity Synthetic Face Generation for Rosacea Skin Condition from Limited Data

TL;DR

This paper explores generating high-fidelity synthetic facial images of Rosacea skin condition using a small dataset and StyleGANs, demonstrating the impact of fine-tuning and regularization on image quality, with expert evaluations and potential for aiding diagnosis.

Contribution

First application of StyleGANs to synthesize high-quality Rosacea facial images from limited data, with analysis of fine-tuning effects and expert validation.

Findings

Fine-tuning and regularization improve image fidelity.

Expert dermatologists validate the synthetic images.

Quantitative metrics support the quality of generated images.

Abstract

Similar to the majority of deep learning applications, diagnosing skin diseases using computer vision and deep learning often requires a large volume of data. However, obtaining sufficient data for particular types of facial skin conditions can be difficult due to privacy concerns. As a result, conditions like Rosacea are often understudied in computer-aided diagnosis. The limited availability of data for facial skin conditions has led to the investigation of alternative methods for computer-aided diagnosis. In recent years, Generative Adversarial Networks (GANs), mainly variants of StyleGANs, have demonstrated promising results in generating synthetic facial images. In this study, for the first time, a small dataset of Rosacea with 300 full-face images is utilized to further investigate the possibility of generating synthetic data. The preliminary experiments show how fine-tuning the model and varying experimental settings significantly affect the fidelity of the Rosacea features. It is demonstrated that $R_1$ Regularization strength helps achieve high-fidelity details. Additionally, this study presents qualitative evaluations of synthetic/generated faces by expert dermatologists and non-specialist participants. The quantitative evaluation is presented using a few validation metric(s). Furthermore a number of limitations and future directions are discussed. Code and generated dataset are available at: \url{https://github.com/thinkercache/stylegan2-ada-pytorch}