Face Detection with YOLOv7: A Comparative Study of YOLO-Based Face Detection Models

Face detection is a subcategory of object detection where the main task is to detect faces in digital media such as images. There is a plethora of face detection models that have been introduced throughout the years that use various deep learning architectures. However, this study focuses on face detection using YOLO architecture. YOLO is a single-stage architecture that is well-known for its accuracy and fast inference speed, making it suitable for various real-time applications. YOLO architecture was introduced back in 2015, and its subsequent generations performed significantly better and became more robust. This study uses the YOLOv7 model as the base model for a face detector and was trained using the Wider Face and FDDB datasets to detect faces. Three variations of face detector modules were created, each with a different number of parameters counts. The lightest model, called YFaces- Tiny, has achieved mAP scores of 94.07%, 92.36%, and 83.15% on the Easy, Medium, and Hard subsets of the Wider Face dataset. YOLOv7face-X, the largest variation, has achieved even better results in all three subsets. Based on the comparative analysis of different face detection YOLO-based models, the YFaces face detector model is capable of competing with other state-of-the-art face detection models.