How it works

1. CaptureUpload an image or take one frame from your camera. Nothing runs continuously.
2. DetectThe server runs a pretrained YOLOv8s model once on the image and returns boxes, class names and scores.
3. FilterSearch by name, pick a category or class, and sort by score, name or box area.
4. FocusSelect an object to highlight it, see a close-up, its 3×3 position and coordinates, or hear its location.

Limitations

Small, hidden, unusual or dim objects may be missed or mislabelled.
Not detected does not mean absent.
The score is the model's confidence, not a guarantee that the label is correct.
Only the 80 COCO object classes can be found. No custom training was done.
FindIt has no accuracy benchmark (no mAP measured). A 20-image feasibility check is in the repository docs.
Images are processed in memory and not stored.

Model details

Stack

Detection: Ultralytics YOLOv8s, pretrained on COCO (80 classes), PyTorch on CPU.
Backend: FastAPI (Python).
Frontend: plain HTML, CSS and JavaScript.

Supported objects

The model can only label these 80 COCO classes. Being on this list does not mean an object will always be found.

Source

FindIt is free software under the GNU Affero General Public License v3.0 (AGPL-3.0), because it uses Ultralytics YOLOv8 (AGPL-3.0). If you use FindIt over a network, you are entitled to its complete source code.

License texts and third-party notices: LICENSE and NOTICE.md in the repository.

Find objects in a photo