Sock Matcher

Phase 1 — image POC

Drop a sock photo here or click to choose a file

Loading models…

Detected sock

Target (you tapped)

Best match

How it works

The app runs entirely in your browser. It uses an open-vocabulary detector (OWLv2) to find socks in the photo, then a vision encoder (DINOv2-small) to compute a feature vector for each sock. When you tap one, it’s compared to every other sock by cosine similarity; the closest is your match.

No images leave your device.