Instance Segmentation

Instance segmentation detects every object instance in an image and produces a separate pixel-level mask for each one.

Unlike semantic segmentation, it distinguishes individual objects within the same class, such as labeling people in a crowd as “person 1,” “person 2,” and so on.

Popular models include Mask R-CNN, YOLACT, and Transformer-based architectures such as Mask2Former.

It is widely used in robotics, retail analytics, medical cell counting, and other tasks that require locating individual objects.