Mask R-CNN simultaneously optimizes three tasks—bounding box regression, classification, and instance segmentation—and is a classic application of multi-task learning in the deep learning era.