Exploring the benefits of heterogeneous computing to accelerate face detection deep learning inference