Think together · Start with a story
A girl wants to find every apple in a photo, not just ask whether there is “some fruit.” The robot needs to point out where they are.
Break down the process
- ClassificationWhat categories are in the image?
- DetectionWhere are the objects?
- SegmentationWhich pixels belong to them?
- Match the taskChoose the needed output
Image classification, object detection, and segmentation are different vision tasks.
Classification gives a category for a whole image or object. Detection often gives a location box, while segmentation marks the pixel area more precisely.
To count apples, a system must consider occlusion, overlap, and missed detections. Seeing the word “apple” is not enough to finish the count.
Your turn to try
If a model says a photo “contains apples,” does it know how many there are?
Ready? Open to see the explanation
No. Checking whether something is present and detecting and counting each item are different tasks.
One idea to remember
Image classification, object detection, and segmentation are different vision tasks.
Try covering the picture above and explaining each step to your family. If one step is hard to explain, revisit that card.