AI Glossary

Vision transformer

Also called ViT

What does “Vision transformer” mean?

Vision transformer: Applies the transformer architecture to images by cutting each picture into small patches and treating them like a sequence of tokens. Given enough training data, vision transformers match or beat convolutional networks on many vision tasks.

Vision transformer visual guide: Applies the transformer architecture to images by cutting each picture into small patches and treating them like a sequence of tokens. Given enough training data, vision transformers match or beat convolutional networks on many vision tasks.
Visual guide Open full size (new tab)

What is an example of “Vision transformer”?

Many vision-language models use a vision transformer as their image encoder, turning a photo into tokens the language model can reason about.

Keep learning

See also

TransformerTermConvolutional neural networkTermVision-language modelTermComputer visionTerm

Same topic

Variational autoencoderTermState-space modelTermSmall language modelTermSelf-attentionTermRecurrent neural networkTerm