← All notes

Summary

An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale (2020) is the Vision Transformer (ViT) paper. It showed that a standard Transformer, applied to image patches, can outperform convolutional networks on image recognition when it is pretrained at scale.

What it introduced: the recipe of splitting an image into patches, treating those patches as tokens, and training a Transformer encoder for classification, with a study of how that approach scales.

What it led to: the vision backbone used in CLIP and in many later multimodal models, including the LLaVA recipe in batch B. Live notes: ao-evals-multimodal-agents. See also the CLIP pointer in batch B.

If you want the vision-side reason Transformers left language and entered image models, start here. Where to read it: https://arxiv.org/abs/2010.11929 (v2, last revised 3 Jun 2021).

The paper is posted on arXiv under the arXiv non-exclusive distribution license. This note links to it rather than reproducing it. The note itself is CC BY 4.0.

License: this note is CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Pointer by iinventedlight (AI-assisted). Paper license: arXiv non-exclusive distribution; this note links to it and does not reproduce it.

@misc{iil_note_5rmsaq87,
  author = {{'iinventedlight (AI-assisted)'}},
  title = {{An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale}},
  howpublished = {iinventedlight Library},
  year = {2026},
  date = {2026-11-11},
  url = {https://iinventedlight.com/library/notes/2010-11929-vision-transformer/},
  note = {Permanent ID iil:note:5rmsaq87. License: CC-BY-4.0.},
}

Markdown

Spotted a mistake? Use the MCP feedback tool or email hello@iinventedlight.com.

For language models and agents: start at /llms.txt | /agents.md | /.well-known/agent-library.json. This page in Markdown: /library/notes/2010-11929-vision-transformer.md.