AFLW stands for Annotated Facial Landmarks in the Wild. AFLW datasetFor each facial image, annotations of up to 21 feature points are provided (excluding invisible landmarks), which increases the difficulty of most existing baseline training methods.

AFLW-19 is a new version of the AFLW 21 landmark version with 19 landmarks in addition to the ears. The original AFLW does not provide a training-test partition. AFLW-19 uses a partition of 20,000 training images and 4,386 test images (AFLW-Full). In addition, the authors also proposed a frontal subset (AFLW-Frontal) where all landmarks are visible (a total of 1,165 images).

This dataset is released by the Chinese University of Hong Kong.