Cross-Age, Gender, and Skin Tone Face Image Dataset

This dataset consists of photorealistic synthetic human face images generated using high-quality image generation models such as StyleGAN and diffusion-based frameworks. It covers a wide range of demographic combinations, including different regions, age groups, genders, and skin tones, providing approximately balanced representation across attributes. The dataset includes children, young adults, middle-aged individuals, and elderly subjects, all rendered with realistic facial detail at a resolution of 1024×1024 pixels. Each image features a centered, unobstructed face under simulated natural lighting, with diverse background styles, natural emotions, and facial expressions.

The dataset is designed for diverse face recognition model training, bias detection, and facial analysis tasks. It is suitable for applications such as generalization testing in face recognition systems, demographic attribute classification, AI fairness evaluation, and multimodal perception system development. Its balanced demographic distribution ensures robustness for training models that can perform well across different population groups and environmental variations.

All images are fully annotated with facial attribute labels, provided in JSON or CSV formats. Labels include age group (child, young, middle-aged, elderly), gender (male, female), and skin tone categories (light, medium, dark, referencing the Fitzpatrick scale). Additional optional attributes can include facial expression type, hair characteristics, and eyewear presence. The combination of multiple attributes across gender × age × skin tone × expression ensures a diverse set of unique individuals and scenes.

The synthetic generation process simulates varied lighting environments and background contexts while preserving consistent facial positioning and clarity. It incorporates subtle variations in facial structure, emotion, and accessory use, enabling rich training data for bias evaluation and cross-domain model adaptation.

By combining photorealistic rendering, controlled attribute diversity, and precise human annotation, this dataset supports the development of high-accuracy, demographically fair face recognition systems and facilitates research into AI bias mitigation across demographic boundaries.

Keywords: face generation diverse face synthetic human virtual face gender age skin tone bias detection recognition

Sector: Biometrics / AI Fairness & Bias Detection / Computer Vision / Multimodal Perception

Metrics: Attribute classification accuracy, bias detection score, cross-group recognition performance

Entities: Male faces, female faces, light-skin faces, medium-skin faces, dark-skin faces, children, young adults, middle-aged adults, elderly individuals

Geographies: Global demographic representation

Price Range (USD): 4000–20000

Source type: Synthetic data generation from simulation environment, human annotation