We provide 1,200 annotated 800x1020 training images in , which part of them are revised from the original “300W” [1]. These images are a subset adapted from the original 300W facial landmark dataset, which have been enhanced to focus solely on facial landmarks. Additional landmarks have been annotated, expanding beyond the standard 68-point annotation scheme to create a comprehensive 51-point facial-landmark detection dataset. This includes a mix of images from the original 300W dataset and our custom images, all annotated with precise facial landmark coordinates.