Explore a trained ViT, one example at a time

Gold compares patch features. Teal shows attention weights. Choose Next example to follow the nine-step tour.

Guided example1. Background finds background2. Check a true corner3. One ear finds the other side4. Move the reference to the nose5. Follow the body’s dark fur6. Rewind to block 17. Ask what the ear reads8. Change only the attention head9. Let CLS gather an image summary PreviousNext example →

Free exploration

Trained ViT · choose a patch to explore

1 · Locate the purple patch

EarNoseTreeCLS

2 · Where does this query read?

Saved pretrained ViT-Tiny · 224×224 RGB · 14×14 patches · 16×16 pixels per selection · Newfoundland: 95.73%