YolPe Demo: Detecting Distant People with Privacy Blur

Upload a photo taken from a high angle, such as a CCTV shot. The model counts the people in it and blurs everyone it finds.

Modelillimax/YolPe (fine-tuned from YOLO26s; about 3.5x better at detecting distant people)
Used In ProductionPowered a live CCTV crowd-monitoring service at the 2026 AI Future Education Expo, hosted by a regional education office in South Korea (3,000+ visitors)
How the Blur WorksEach detected person is downscaled to 16 px on the long side, then scaled back up. Faces end up only about 2 px wide, so they can’t be recovered.
Service Codelkw-k/YOBA · Project page

Everything runs right in your browser. Your photos are never uploaded or stored. The model (about 39 MB) is downloaded once, the first time you run it.

Loading the model...
Detections
Blurred output (what the service receives)