Three different questions to ask an image
Classification answers 'what is the main thing in this image': feed in a photo, get back a single label like 'golden retriever' with a confidence score, this is what powers apps that identify a plant species from a photo. Object detection goes further and answers 'what things are in this image and where,' drawing a bounding box around each object it finds and labeling it, which is what lets a self-driving car's perception system say 'pedestrian at these coordinates, car at those coordinates' in the same frame.
Segmentation goes further still and answers 'exactly which pixels belong to which object,' producing a precise outline rather than a rough box. Semantic segmentation labels every pixel by category, like 'road,' 'sky,' 'building,' while instance segmentation goes a step further and separates individual objects of the same category, so two overlapping people each get their own distinct pixel mask instead of being merged into one 'person' blob.