.png)
Folder Person Search
Search a server-side image folder for a specific individual using a reference photo
Folder Person Search scans an entire folder of images on the server and finds every photo in which a specific person appears. Instead of searching video frames in real time, it operates on a pre-existing image archive — useful for reviewing HR records, event photo libraries, or CCTV snapshots.
How It Works
- Upload a reference photo — the face in the image is embedded using InsightFace ArcFace (512-d cosine embedding).
- Select a server folder — choose from the list of image directories indexed on the backend.
- Run the search — the system iterates over every image in the folder, detects faces, and computes the similarity between each detected face and the reference embedding.
- Review results — matched images are returned ranked by similarity score, with the file path and confidence displayed.
Similarity Threshold
The similarity threshold controls how strict the matching is:
| Threshold | Behaviour |
|---|---|
0.6 – 0.7 | High precision — only very close matches returned |
0.45 – 0.55 | Balanced — recommended for most use cases |
0.3 – 0.4 | High recall — more results, higher chance of false positives |
Adjust the threshold slider before running the search. The default is 0.5.
Server Folder Configuration
Folders are indexed on the backend and exposed via the /api/analyze/images/server-folders endpoint. To add a new folder to the list, configure the backend's SEARCH_FOLDERS setting to include the target directory path.
Webcam Alternative
If you don't have a reference photo available, use Webcam Folder Search to capture the reference face live from a webcam instead.
Best Practices
- Use a clear, front-facing reference photo — avoid sunglasses, heavy blur, or extreme angles.
- For large folders (500+ images), a GPU-accelerated backend significantly reduces search time.
- The search is non-destructive — no images are modified or moved.