I don't think the only utility of a depth model is to provide synthetic blurring of backgrounds. There are many things you'd like to use them for, including feeding into object detection pipelines.