Instance Segmentation identifies object boundaries and labels each pixel in an image with both class and instance information. It differs from semantic segmentation, which only labels classes without distinguishing individual instances. Advanced architectures like Mask R-CNN excel at this task.