High-Resolution Face Editing using Pre-Trained GAN and Automated Evaluation Framework for Quality Assessment

High-resolution face editing has emerged as a key focus within the domain of computer vision, with applications spanning across various domains. This paper presents a novel framework that combines the power of pre-trained generative adversarial networks (GANs), specifically StyleGAN2-ADA, with an automated evaluation system for image quality assessment. The approach consists of three stages: (i) generating high-resolution facial images using a pre-trained GAN model, eliminating the need for custom training datasets; (ii) editing selected facial attributes in the generated images to achieve realistic modifications while preserving the overall facial structure, and displaying the original and edited images side by side for comparison; and (iii) employing an automated evaluation framework that analyzes the edited images for quality using measurements like Signal-to-Noise Ratio (SNR) or Peak Signalto- Noise Ratio (PSNR). The proposed method emphasizes both realism and precision in the facial edits, alongside a seamless and objective quality evaluation mechanism.