The paper introduces the WIDER FACE dataset, a large-scale face detection benchmark that is 10 times larger than existing datasets. The dataset contains 32,203 images with 393,703 labeled faces, featuring rich annotations such as occlusions, poses, event categories, and face bounding boxes. The faces in the dataset are highly challenging due to variations in scale, pose, and occlusion. The authors benchmark several state-of-the-art detection systems and propose a multi-scale two-stage cascade framework to address large-scale variations. They also demonstrate that WIDER FACE is an effective training source for face detection, showing significant improvements in performance when retrained on this dataset. The paper discusses common failure cases and highlights the need for further research in handling small-scale, occluded, and extreme pose faces, which are crucial for real-world applications.The paper introduces the WIDER FACE dataset, a large-scale face detection benchmark that is 10 times larger than existing datasets. The dataset contains 32,203 images with 393,703 labeled faces, featuring rich annotations such as occlusions, poses, event categories, and face bounding boxes. The faces in the dataset are highly challenging due to variations in scale, pose, and occlusion. The authors benchmark several state-of-the-art detection systems and propose a multi-scale two-stage cascade framework to address large-scale variations. They also demonstrate that WIDER FACE is an effective training source for face detection, showing significant improvements in performance when retrained on this dataset. The paper discusses common failure cases and highlights the need for further research in handling small-scale, occluded, and extreme pose faces, which are crucial for real-world applications.