DinoLens · Correspondence

Hover on one image — see where the same feature lives in the other. Used in similar image search

Loading model…
1.00x
A Source Image

Drop image or click to upload

PNG, JPG, WebP

or

B Target Image

Drop image or click to upload

PNG, JPG, WebP

or

💡 How it works: DINOv3 splits each image into small patches and computes a feature vector for each. When you hover a patch on one image, we find the patch in the other image with the highest cosine similarity — revealing that DINOv3 understands semantic correspondence (e.g., "left eye" maps to "left eye") without any labels.