Image Translation Between Human Faces and Wayang Orang using U-GAT-IT

This study presents the application of U-GAT-IT, a state-of-the-art generative adversarial network, for unpaired image-to-image translation between human facial images and Wayang Orang, a traditional Javanese theatrical art form known for its intricate and vibrant stylistic features. The proposed framework leverages U-GAT-IT’s attention mechanisms and Adaptive Layer-Instance Normalization (AdaLIN) to effectively balance the preservation of human identity features with the detailed artistic elements of Wayang Orang. A dataset comprising 500 images from each domain was curated to train the model using adversarial, cycle-consistency, and identity losses. The model was optimized using the Adam optimizer and evaluated using Frechet Inception Distance (FID) and Structural Similarity Index (SSIM). The results demonstrate high-quality and stylistically faithful translations, with FID scores of 23.87 (human-to-Wayang Orang) and 25.12 (Wayang Orang-to-human), and SSIM scores of 0.84 and 0.81, respectively. Qualitative evaluation confirmed the model’s ability to generalize to unseen images while maintaining content fidelity and stylistic richness. This research highlights the potential of generative models in digital cultural preservation and opens new avenues for applications in virtual storytelling, education, and creative industries. Future work will explore real-time deployment and expansion to other culturally rich artistic domains.