CelebA

CelebA, short for CelebFaces Attributes Dataset, is a large-scale collection of celebrity face images that has become a standard benchmark in computer vision for facial attribute analysis and related tasks. The dataset comprises more than 200,000 images of about 10,000 celebrities. Each image is annotated with 40 binary attributes (for example, gender, age group, smiling, wearing glasses, and various hair and makeup features) and five annotated facial landmarks. In addition, each image includes an identity label, and images are aligned and cropped to a common size to facilitate processing. The five landmark points correspond to the left eye, right eye, nose tip, left mouth corner, and right mouth corner.

The data are organized to support a range of research tasks. Standard practice provides training, validation, and test subsets. CelebA is widely used for attribute prediction and multi-label classification, face verification, and landmark localization. It also serves as a resource for attribute-guided image generation, editing, and other generative model studies, where attribute vectors influence synthesis or transformation of facial images. The combination of a large number of identities, substantial image count, rich attribute annotations, and precise landmarks has made CelebA a practical benchmark for evaluating robustness to variations in pose, lighting, and background.

CelebA has had a broad impact on the computer vision and machine learning communities, supporting both foundational research in facial attributes and applied developments in face editing, recognition, and synthesis. The dataset remains a commonly referenced resource for researchers seeking a scalable, richly annotated facial attribute dataset.