MAAD-Face: A Massively Annotated Attribute Dataset for Face Images

Soft-biometrics play an important role in face biometrics and related fields since these might lead to biased performances, threaten the user's privacy, or are valuable for commercial aspects. Current databases specifically constructed development of recognition applications. Consequently, contain a large number images but lack attribute annotations overall annotation correctness. In this work, we propose novel annotation-transfer pipeline that allows accurately transfer from multiple source datasets target dataset. The is based on massive classifier can state its prediction confidence. Using confidences, high correctness transferred ensured. Applying VGGFace2 database, MAAD-Face database. It consists 3.3M faces over 9k individuals provides 123.9M 47 different binary attributes. it 15 137 times more than CelebA LFW. Our investigation quality by three human evaluators demonstrated superiority existing databases. Additionally, make use high-quality study viability soft-biometrics recognition, providing insights into which attributes support genuine imposter decisions. dataset publicly available.