The purpose of instance segmentation is to segment the mask of each object instance from the images. Compared with semantic segmentation, instance segmentation not only distinguishes different categories, but also distinguishes different object instances within the same category. As shown in Fig. 13.1, in the results of semantic segmentation, the labels corresponding to different sheep are the same; that is, they are all assigned to the category of “sheep”, while in the results of instance segmentation, the category labels of different sheep are the same, but they have different instance numbers; compared with object detection, instance segmentation needs to further segment the mask of the object within each bounding box.