ebowwa/usd-side-coco-annotations

USD Side Detection Dataset (Front/Back)

A refined COCO-format dataset for detecting US Dollar currency and classifying whether the front or back side is visible.

Dataset Summary

• Total Images: 7,360
• Format: COCO (object detection)
• Classes: 33 (denominations × front/back × authentic/counterfeit)

Split	Images
Train	5,290
Valid	1,221
Test	849

Classes

Denomination + Side (18 classes)

• 100USD-Front, 100USD-Back
• 50USD-Front, 50USD-Back
• 20USD-Front, 20USD-Back
• 10USD-Front, 10USD-Back
• 5USD-Front, 5USD-Back
• 1USD-Front, 1USD-Back

Counterfeit Detection (15 classes)

• Counterfeit versions for each denomination

Annotation Refinement

This dataset was refined using Roboflow's usd-classification/1 model to reclassify generic labels (e.g., 100USD) into specific front/back variants:

• 2,236 annotations auto-reclassified
• 97% success rate on classification

Usage

from datasets import load_dataset

dataset = load_dataset("ebowwa/usd-side-coco-annotations")

Or download the zip directly and extract for use with YOLO/RF-DETR training.

Source

Original dataset from Roboflow - "Front/Back of USD" project.

License

MIT