Vision Transformer
Definition
A Vision Transformer processes images as sequences of small image patches using a Transformer architecture.
An image model architecture that processes image patches as a token sequence with a Transformer.
Simply explained
The AI breaks an image into small tiles and considers their relationships to understand the whole image.
Why it matters in practice
The term helps to select, develop, evaluate AI systems in a well-founded manner and realistically assess their technical risks.
Example from business
A quality inspector compares ViT and CNN on defects with limited training data.
Limits and pitfalls
- ViT is not a single fixed architecture; data requirements, patch size, and pretraining influence performance and cost.
Origin & history
Dosovitskiy and coauthors published ViT in 2020 and showed strong results with large-scale pretraining.
Use in mid-sized companies
- Possible application: A quality inspector compares ViT and CNN on defects with limited training data.
Frequently asked questions
What does Vision Transformer mean?
An image model architecture that processes image patches as a token sequence with a Transformer. What should you watch out for? ViT is not a single fixed architecture; data requirements, patch size, and pretraining influence performance and cost.
Sources
- An Image is Worth 16x16 Words · Dosovitskiy et al., An Image is Worth 16x16 Words
Source checked on 10.10.2026. Editorial application examples are hypothetical; no evidence of real customer results.
Editorially reviewed on: