Semantic segmentation: every pixel gets a class
Semantic segmentation labels every single pixel in an image with a class, like 'road', 'sky', 'building', or 'person'. It answers 'what is at this pixel' for the entire image. Crucially, semantic segmentation does not distinguish between separate objects of the same class. If a photo has three people standing together, all the pixels belonging to any of them just get labeled 'person', and they blend into one connected blob rather than three separate shapes.
That's fine for classes like 'sky' or 'road', where you genuinely don't care about counting individual instances, there's just one sky. But it's a real limitation for classes made up of distinct individual objects, where knowing 'this is a person' isn't as useful as knowing 'this is person A, and that is a separate person B'.