Instance Segmentation

Instance segmentation identifies every individual object in an image and assigns a precise pixel-level mask to each one. It combines the goals of object detection, which locates separate items, and semantic segmentation, which labels pixels by category. A street scene system can therefore distinguish three different cars rather than marking all car pixels as one region. Applications include autonomous navigation, medical imaging, robotics, agriculture, retail analysis, and industrial inspection. Models require detailed training labels and can struggle with overlapping objects, unusual viewpoints, poor lighting, or very small targets. Evaluation considers both whether each instance was detected and how accurately its mask matches the object’s visible boundaries.