Latent Space Exploration of StyleGAN2-ADA for Fashion Image Synthesis: A Practical Implementation Guide for Moderate-Sized Datasets

Authors

• Ali H. Shareef, Research Groups in Intelligent Machines (REGIM Lab), University of Sfax, National Engineering School of Sfax (ENIS), BP 1173 Sfax, 3038, Tunisia, National School of Electronics and Telecommunications of Sfax (ENET'Com), University of Sfax, Sfax, 10587, Tunisia; AND University of Thi-Qar,Nasiriyah, 64001, Thi-Qar, IraqFollow
• Hajer Ghodhbani, Research Groups in Intelligent Machines (REGIM Lab), University of Sfax, National Engineering School of Sfax (ENIS), BP 1173 Sfax, 3038, Tunisia AND Department of Electrical and Electronic Engineering Science, University of Johannesburg, Auckland Park, 2006, Johannesburg, South Africa
• Tarek M. Hamdani, Research Groups in Intelligent Machines (REGIM Lab), University of Sfax, National Engineering School of Sfax (ENIS), BP 1173 Sfax, 3038, Tunisia AND Higher Institute of Computer Science Mahdia, University of Monastir, Mahdia, 5111, Tunisia
• Habib Chabchoub, College of Business, Al Ain University of Science and Technology, Abu Dhabi, United Arab Emirates
• Adel M. Alimi, Research Groups in Intelligent Machines (REGIM Lab), University of Sfax, National Engineering School of Sfax (ENIS), BP 1173 Sfax, 3038, Tunisia AND Department of Electrical and Electronic Engineering Science, University of Johannesburg, Auckland Park, 2006, Johannesburg, South Africa

Abstract

\hlThis paper presents the use of StyleGAN2-ADA for fashion image synthesis and editing. We demonstrate that generative adversarial networks (GANs) can generate realistic and diverse fashion images even under moderate data constraints. The model trained on a dataset of 9,000 fashion images achieved a Fréchet Inception Distance (FID) of 10.8, a Garment Structure Accuracy (GSA) of 0.85, a Pattern Continuity Score (PCS) of 0.78, and an Attribute Transfer Accuracy (ATA) averaging 75% across key attributes. These quantitative results confirm the model’s ability to produce high-quality, structurally consistent, and semantically coherent fashion images. Unlike previous studies focusing on either extremely limited or very large datasets, our work explicitly addresses the research gap in practical, moderate-sized scenarios where computational resources are constrained but data availability is not severely scarce. The study also provides practical implementation guidelines for researchers and fashion designers working with limited resources. The source code developed and used in this study is publicly available on GitHub at the following link: GitHub Repository.

Recommended Citation

Shareef, Ali H.; Ghodhbani, Hajer; Hamdani, Tarek M.; Chabchoub, Habib; and Alimi, Adel M. (2026) "Latent Space Exploration of StyleGAN2-ADA for Fashion Image Synthesis: A Practical Implementation Guide for Moderate-Sized Datasets," Iraqi Journal for Computer Science and Mathematics: Vol. 7: Iss. 2, Article 13.
DOI: https://doi.org/10.52866/2788-7421.1413
Available at: https://ijcsm.researchcommons.org/ijcsm/vol7/iss2/13