The database systematically varies images across these categories:
: Subjects are photographed wearing different accessories, including various pairs of glasses and different types of caps. Cas Peal
: The full database includes 99,594 images, while a publicly available subset called CAS-PEAL-R1 contains 30,900 images. : The dataset provides metadata such as eye
: Images are captured from multiple angles using a horizontal semicircular camera array to simulate different head rotations. Core "PEAL" Features is a large-scale Chinese face
: The dataset provides metadata such as eye positions, face pose angles, and expression labels to help refine detection systems.
The database was developed by the Joint Research & Development Laboratory (JDL) of the Chinese Academy of Sciences (CAS). Its primary purpose is to provide researchers with a robust dataset for training and testing algorithms against real-world visual challenges. Core "PEAL" Features
is a large-scale Chinese face database created to evaluate face recognition algorithms, specifically focusing on four key types of variations known as PEAL : P ose, E xpression, A ccessories, and L ighting .