Recent advancements in deep learning-based face aging and de-aging have been remarkable. However, existing methods have two primary limitations. First, widely used high-resolution datasets, such as FFHQ, are predominantly composed of Caucasian subject...
Recent advancements in deep learning-based face aging and de-aging have been remarkable. However, existing methods have two primary limitations. First, widely used high-resolution datasets, such as FFHQ, are predominantly composed of Caucasian subjects, leading to racial bias that fails to accurately reflect the skin characteristics and aging signs of East Asians. Second, global style transfer approaches based on StyleGAN often result in the blurring of high-frequency textures, such as wrinkles and pores, or cause subtle distortions in the subject's identity. In this paper, we propose a SPADE-based high-resolution (512x512) conditional GAN (cGAN) model capable of independently controlling wrinkles, pores, and redness by utilizing Skin Condition Maps optimized for East Asian facial characteristics. To achieve this, we constructed a refined dataset based on East Asian images from AI-Hub and FFHQ and introduced a skin masking strategy along with Cycle Consistency Loss for precise control in an unpaired setting. Experimental results demonstrate that the proposed model robustly preserves identity while enabling precise generation and control of skin micro-features compared to existing methods.