“Do you know what segmentation is?” Jon asked the magician. The magician replied: “It’s the process of dividing an image into meaningful regions”
“like giving every pixel a role. So, you know the pixels that correspond to a cat, the pixels that belong to the tree, and so on…They are the reasons why in image editing apps, you can crop the exact part that contains the cat”
“What if I want to know not just “cat”—but “Cat #1,” “Cat #2,” and “Cat that knocked over the vase.”?”
“That’s Instance Segmentation. You want to know not only the type, but also the instance 1, instance 2,…Meanwhile, semantic segmentation labels every pixel like “grass,” “sky,” etc.”
“And I heard there is also Panoptic Segmentation. It does both”
“Really?”