Cut an image into patches and treat them like words in a transformer.
- PatchesSplit a 224x224 image into 16x16 patches.
- EmbedEach patch becomes a vector plus position.
- TransformerPatches attend to each other.
- ClassifyA class token gives the label.
Explain 1/5
Source: Dosovitskiy et al.: An Image is Worth 16x16 Words