"FFHQ"

"The Flickr-Faces-HQ (FFHQ) dataset was originally established as a benchmark for Generative Adversarial Networks (GANs) and served as the primary training data for StyleGAN. It was open-sourced by NVIDIA in 2019. FFHQ is a high-quality dataset comprising 70,000 high-resolution PNG images at 1024\u00d71024 resolution. It is characterized by rich diversity in age, ethnicity, and image backgrounds. The dataset exhibits extensive variation in facial attributes, including age, gender, ethnicity, skin tone, expression, face shape, hairstyle, and pose. It also includes a wide array of accessories such as eyeglasses, sunglasses, hats, headwear, and scarves. Consequently, FFHQ is highly suitable for developing models for facial attribute classification and semantic segmentation. The images were crawled from Flickr, ensuring that all downloads complied with licensing agreements. The data was processed using dlib for face alignment and cropping, and algorithms were employed to filter out non-real faces, such as statues, paintings, and photographs of photos."