Currency Vision

Live denomination detector

Currency Vision

Detect banknote denominations from a photo. The image stays in your browser; the app draws boxes and totals the money it finds.

36 denominations 6 currencies Browser inference
View benchmark Open paper

Detector

Try a photo

local image processing
Samples
Choose an image to load the detector.

No image selected

Benchmark

Five detector architectures, one banknote dataset

The dataset contains 3,793 smartphone photos and 4,320 annotated banknotes across USD, PHP, INR, EUR, AUD, and CAD.

Architecture Precision Recall mAP@0.5 mAP@0.5:0.95
YOLOv8s running above 0.9310.8780.9170.780
YOLOv8-SE 0.6210.6700.7050.553
YOLOv8m best mAP@0.5:0.95 0.9540.9000.9350.812
RT-DETR 0.9370.9170.9370.811
CoordAtt + AdaptiveSPPF proposed 0.8960.8580.9020.771
Detected banknotes with bounding boxes

Proposed model training curves

mAP@0.5 training curve mAP@0.5:0.95 training curve Precision training curve Recall training curve
Full paper Source repository Dataset