Litter detection demo
A small detector (YOLO11s or YOLO11n) finds cigarette butts on pavement. It runs entirely in your browser, on your GPU where the browser supports WebGPU; photos are never uploaded anywhere.
Try an example
Validation photos from the same walk; none of them was used in training. Each one is about 2 MB.
78% and up 65–78% below 65%
Pick an example or upload a photo. Each model (11–38 MB) loads on first use.
Every detection is shown with the model's score, coloured by score: green at 78% and up, orange at 65–78%, red below 65%. Lower scores hold more of the false alarms. Switching the model runs detection again on the same photo. With a mouse, hovering over the photo shows it magnified 2× and 6× at full resolution.