They are using deep learning and also other sensors may contribute to the depth map (assuming they have one).