ImageGBT 1.5 Detector (Minimal)
Frozen backbone detector for ImageGBT 1.5 (95-97% accuracy)
Frozen backbone transfer learning approach:
- ❄️ Pretrained ViT backbone (frozen)
- ✅ Trained classifier head only
- 📉 Minimal augmentation
- 🎯 Optimized for 400 images
Performance
| Metric | Score |
|---|---|
| Accuracy | 0.9750 (97.50%) |
| Precision | 0.9524 |
| Recall | 1.0000 |
| F1 Score | 0.9756 |