SoF stands for Specs on Faces. It is a dataset used to evaluate face detection, recognition and classification technologies. It helps to overcome the challenges brought by lighting environment and face occlusion in this field. The dataset collects 42,592 images of 112 volunteers (66 men and 46 women) wearing glasses under different lighting conditions. The SoF dataset is mainly designed to solve the two major challenges of face detection, recognition and classification, namely harsh lighting environments and face occlusion.