Update Readme.md
Browse files
README.md
CHANGED
|
@@ -28,11 +28,24 @@ This model is a fine-tuned version of **Meta's DINOv3-ViT-S/16** for binary clas
|
|
| 28 |
|
| 29 |
The model was developed to provide a high-reliability baseline for separating formal block scripts (**Uchen**) from cursive script families (**Ume**). By focusing on global page geometry rather than local character patches, it achieves high accuracy on whole-page manuscript scans.
|
| 30 |
|
| 31 |
-
- **Developed by:** OpenPecha / [Your Name]
|
| 32 |
- **Model type:** Vision Transformer (ViT)
|
| 33 |
- **Language(s):** Tibetan (Classical/Manuscript)
|
| 34 |
- **Finetuned from model:** facebook/dinov3-vits16-pretrain-lvd1689m
|
| 35 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 36 |
## Uses
|
| 37 |
|
| 38 |
### Direct Use
|
|
|
|
| 28 |
|
| 29 |
The model was developed to provide a high-reliability baseline for separating formal block scripts (**Uchen**) from cursive script families (**Ume**). By focusing on global page geometry rather than local character patches, it achieves high accuracy on whole-page manuscript scans.
|
| 30 |
|
|
|
|
| 31 |
- **Model type:** Vision Transformer (ViT)
|
| 32 |
- **Language(s):** Tibetan (Classical/Manuscript)
|
| 33 |
- **Finetuned from model:** facebook/dinov3-vits16-pretrain-lvd1689m
|
| 34 |
|
| 35 |
+
## Performance Summary
|
| 36 |
+
The model achieved near-perfect discrimination during testing, specifically excelling in identifying formal Uchen with high recall.
|
| 37 |
+
|
| 38 |
+
- **Test Accuracy:** 98.95%
|
| 39 |
+
- **Macro F1-Score:** 0.984
|
| 40 |
+
- **AUC-ROC:** 0.9988
|
| 41 |
+
- **Best Training Configuration:** Stage B (Partial Backbone Unfreezing)
|
| 42 |
+
|
| 43 |
+
### Confusion Matrix
|
| 44 |
+
| Predicted \ Actual | Uchen | Ume |
|
| 45 |
+
|--------------------|-------|-----|
|
| 46 |
+
| **Uchen** | 159 | 2 |
|
| 47 |
+
| **Ume** | 6 | 595 |
|
| 48 |
+
|
| 49 |
## Uses
|
| 50 |
|
| 51 |
### Direct Use
|