Face detection is a crucial step for several applications including surveillance, human-machine interaction, and IoT. Robustness to occlusion, pose, scale, and illumination changes are the key issues in all of these systems. After the success of CNNs in object detection, face detection has been dominated by the CNN-based methods. This paper proposes a face detector designed based on a recently introduced real time deep object detector, YOLOv3. In particular YOLOv3 network is trained as a face detector and a new model file is generated. Performance evaluation reported on WIDER data base demonstrate that the developed face detector, YOLOv3-face, improves robustness to occlusion and pose changes and it is capable of detecting faces greater than 15 pixels. Performance of YOLOv3-face compared to top 14 state-of-the-art trackers is reported in terms of precision-recall curves. It is concluded that the precision rates achieved by YOLOv3-face are very close to the top 8 trackers at tolerable false detection rates. Moreover, the computational load of YOLOv3-face detector significantly reduces with the used single pass joint optimization scheme.