# Face Sorter

> Group a folder of your own photos by person, entirely in the browser. Face detection and face matching run on
> the device; photos are never uploaded and faces are never compared against any outside database.

URL: https://face-sorter.skillsafe.ai/
Price: free, no sign-up, no account.

## What it does
- Pick a folder (or several photos, or drop them in). Every image is decoded locally.
- Finds faces with YuNet (whole photo plus overlapping tiles, so small faces in group photos are found too).
- Aligns each face on five landmarks and describes it with an on-device face-matching model (SFace), then groups
  faces by person with average-linkage clustering. A slider makes grouping stricter or looser.
- You name people, merge groups ("Maybe the same person?" suggestions included), split groups by moving selected
  faces to a new or existing person, send faces back to "unsorted", or remove non-faces.
- Exports: CSV (file, person, face box, confidence; photos with no face are listed too), JSON (people -> files,
  files -> people), and a zip with one folder per person (or one person's zip). Zips are built in the browser.
- "See an example" sorts ten public-domain NASA photos of the Apollo 11 crew at no cost.

## Privacy
Nothing leaves the browser tab. The only network traffic is the app itself and the model files (about 27 MB of
WebAssembly runtime, a 0.2 MB detector and a 10 MB face-matching model, about 37 MB in all, then cached). There is no server-side processing, no upload and no
face database: faces are only compared with other faces in the set the user picked.

## If the face-matching model cannot load
Automatic grouping uses the SFace model served by SkillSafe's model host. If it cannot be reached, the app still
finds every face and offers manual sorting, and the example falls back to face vectors computed in advance with the
same pipeline.

## Models and licences
- YuNet 2023mar face detector, OpenCV Zoo, MIT.
- SFace (MobileFaceNet, SFace loss) int8 face embedder, OpenCV Zoo, Apache-2.0.
- onnxruntime-web 1.27 (WebAssembly), MIT.
- Example photos: NASA, public domain, via Wikimedia Commons.