Semantic segmentation is a type of computer vision technique that assigns a label to every pixel in an image. The label indicates the class of object that the pixel represents. Semantic segmentation is used in tasks such as self-driving cars, where it is important to know not only the boundaries of objects, but also what those objects are.