A high-precision (92.3%!) small yolov8 model trained to detect all kinds of worn rings.

If you want a global jewelry detection model, check this out!

📊 Performance Metrics:

• Precision (B): 0.923

• Recall (B): 0.911

• mAP@50: 0.948

• mAP@50-95: 0.661

ℹ️ Usage:

Recommend using a bbox threshold of 0.50 - 0.60.

❌ Limitations:

1. Bracelets and watches tend to cause false positives

2. Eyes can cause false positives (rare)

3. It works with anime style but with less confidence.