Caltech 10k Web Faces Dataset is a dataset composed of images obtained from Google searches. The center coordinates of the eyes, nose, and mouth on the front of each image are provided in the relevant files. The relevant information can be used to align and crop faces and benchmark facial recognition algorithms. The dataset has a total of 10,524 images of different resolutions and is used in different fields.

This dataset was released by California Institute of Technology in 2005, and the related paper is "Pruning Training Sets for Learning of Object Categories".