Index Ai Academy
Glossary · Phase 5 · Transformers

Instance segmentation

Giving every pixel a class and an object, so that two touching cars stay two cars.

Try it now

Learn it properly

Segmentation: a label for every pixel · Masks, and how to grade them with IoU and Dice; why a pixel's label needs context and too much context blurs; the shrink-then-enlarge network with skip connections; and turning a mask into countable objects. About 45 minutes.

Other terms from the same lesson

  • connected components: The separate groups of touching pixels in a mask, found by spreading out from each unvisited pixel.
  • Dice score: A score for how well two masks agree: twice the pixels they share, divided by their two sizes added together.
  • receptive field: The part of the input image that one output of a network can see.
  • segmentation mask: An array the size of an image holding one class number per pixel.
  • semantic segmentation: Giving every pixel of an image a class, without separating one object of a class from the next.
  • U-Net: A segmentation network that shrinks the image to understand it, enlarges it again to answer per pixel, and passes fine detail across with skip connections.
Keep learning

More in plain English.

1286 terms in the glossary, each with the lesson that teaches it.

1286 terms

The glossary

Every word you'll meet in AI and machine learning, explained in everyday language.

Phase 5 · lesson 45 min

Segmentation: a label for every pixel

Masks, and how to grade them with IoU and Dice; why a pixel's label needs context and too much context blurs; the shrink-then-enlarge network with skip connections; and turning a mask into countable objects.

Never trained an AI? Teach a small one to write, right in your browser. It takes about 20 seconds and needs no account. Teach it to write