Instance segmentation is a computer vision task that identifies and segments each individual object in an image at the pixel level. Unlike semantic segmentation—which labels pixels by class only—instance segmentation distinguishes between different objects of the same class (e.g., separating two dogs in the same image).

It combines object detection (locating and classifying objects with bounding boxes) with pixel-wise segmentation. Popular models include Mask R-CNN, YOLACT, and Detectron2, which build on object detection frameworks by adding a mask prediction branch for each detected object.

Instance segmentation is used in applications like autonomous driving, robotics, medical imaging, and image editing, where precise object boundaries are important. The output is typically a set of masks, each corresponding to one detected instance, along with class labels and confidence scores.

Challenges include overlapping objects, fine boundary detection, and balancing accuracy with real-time performance, especially in deployment on edge devices.