U-Net

← Back to the article

U-Net finds exact shapes in an image by shrinking the image down and then building a detailed map back up.

What it does

U-Net finds exact shapes in an image by shrinking the image down and then building a detailed map back up.

Segmentation: Coloring specific pixels • Architecture: Brain shape

How it works

  • Shrink the image to understand the big picture.
  • Grow it back to the original size.
  • Mark which pixels belong to the object.

Uses and applications

  • A medical tool can outline organs or tumors for a doctor to inspect.
  • A map tool can outline roads or buildings in satellite images.
  • A photo editor can separate a person from the background.

In short

U-Net is useful for pixel-level image masks.

Related methods

  • Mask R-CNN: Produces a mask per detected object instead of one image-wide map.
  • Diffusion: Many diffusion models use a U-Net to predict and remove noise.
  • ResNet: Also passes information through skip connections.

Back to Computer vision · All topics

The interactive demonstration and the Python walkthrough for this topic need JavaScript.