Vision Transformer (ViT)

← Back to the article

A Vision Transformer cuts a picture into small squares and checks how the squares relate to each other.

What it does

A Vision Transformer cuts a picture into small squares and checks how the squares relate to each other.

Patch: A small square of a photo • Attention: Focusing on what matters

How it works

  • Cut the image into patches.
  • Treat each patch like a word in a sentence.
  • Use attention to connect the patches that matter.

Uses and applications

  • A photo app can understand a whole scene, not just one corner.
  • A medical tool can compare areas of a scan.
  • A satellite tool can study land, roads, or buildings in large images.

In short

It treats an image like a puzzle to understand the big picture.

Related methods

  • Transformer: The language architecture that ViT adapts to images.
  • ResNet: A convolutional approach to image classification.
  • CLIP: Commonly uses a ViT to encode images.

Back to Computer vision · All topics

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