Instance segmentation

Instance segmentation: The computer vision task that delivers pixel-level masks for every distinct object, even when they share a class.

This technology combines object detection with semantic segmentation, going beyond simple bounding boxes to provide a unique binary mask for each object instance. Unlike semantic segmentation, which labels all 'car' pixels as one group, instance segmentation assigns a separate ID and mask to 'Car 1,' 'Car 2,' and 'Car 3,' even if they overlap. Key models like Mask R-CNN and YOLOv8 drive this precision. Its primary value is in applications demanding granular detail: autonomous vehicles (differentiating pedestrians), medical imaging (isolating individual cells or tumors), and quality control (identifying specific product defects on a line).

https://blog.roboflow.com/what-is-instance-segmentation/