Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
|
@@ -8,7 +8,7 @@ tags:
|
|
| 8 |
- edge-detection
|
| 9 |
---
|
| 10 |
|
| 11 |
-
# Deep Tattoo Segmentation v5
|
| 12 |
|
| 13 |
Edge-Aware Attention U-Net for precise tattoo extraction with transparent background.
|
| 14 |
|
|
@@ -20,6 +20,12 @@ Edge-Aware Attention U-Net for precise tattoo extraction with transparent backgr
|
|
| 20 |
- Trained on 24 manual labels + 165 auto-generated masks
|
| 21 |
- Test-Time Augmentation (TTA) support
|
| 22 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 23 |
- `edge_aware_improved_best.pth`: Base Model (Val Dice: 79.17%)
|
| 24 |
- Foundation model before hybrid training
|
| 25 |
|
|
|
|
| 8 |
- edge-detection
|
| 9 |
---
|
| 10 |
|
| 11 |
+
# Deep Tattoo Segmentation v5/v7
|
| 12 |
|
| 13 |
Edge-Aware Attention U-Net for precise tattoo extraction with transparent background.
|
| 14 |
|
|
|
|
| 20 |
- Trained on 24 manual labels + 165 auto-generated masks
|
| 21 |
- Test-Time Augmentation (TTA) support
|
| 22 |
|
| 23 |
+
- `edge_aware_v7_samrefiner_best.pth`: **v7 Model with SAMRefiner** (Val Dice: 74.25%)
|
| 24 |
+
- Same architecture as v5
|
| 25 |
+
- Trained with 122 SAMRefiner-refined masks + 25 v2 fallback
|
| 26 |
+
- SAMRefiner (ICLR 2025) for mask refinement
|
| 27 |
+
- Better edge quality but lower validation score (due to v2-only validation set)
|
| 28 |
+
|
| 29 |
- `edge_aware_improved_best.pth`: Base Model (Val Dice: 79.17%)
|
| 30 |
- Foundation model before hybrid training
|
| 31 |
|