Training a YOLOv2 model with the Darknet object detection framework for real-time detection of playing cards
Click the video below:
I ran YOLOv3 on my own generated dataset of 60,000 images (20% saved as test set). Results:
| mAP | IoU | F1-Score |
|---|---|---|
| 99.86% | 82.92% | 0.99 |
The script runs inference at real-time (30fps) on a live video feed.
- Only tested on Bicycle cards. I chose this deck due to its popularity in commercial use.