Upload unet3plus+efficientnet model, code, and model card

README.md

@@ -56,6 +56,7 @@ Evaluated on the fixed 53-image test partition of Kvasir-SEG (50 % of the origin
 | Precision | 0.9474 |
 | Recall | 0.9005 |
 | Accuracy | 0.9745 |
+| Loss | 0.0914 |
 
 ## Comparison with Sweep Models
 
@@ -95,7 +96,7 @@ The best trial (#32) achieved `eval_loss = 0.0537` (target: < 0.10 ✓).
 - **Optimiser:** AdamW
 - **Epochs:** 50 (sweep) + 5 (HPO final retrain)
 - **FP16:** enabled
-- **Dataset:** Kvasir-SEG augmented (5,368 train / 53 val / 53 test)
+- **Dataset:** Kvasir-SEG augmented (4,800 train / 100 val / 100 test)
 - **Augmentation:** random H/V flips, ±30° rotation, brightness/contrast/saturation ±20 %
 
 ## How to Use

config.json

@@ -18,12 +18,13 @@
     "optimiser": "AdamW"
   },
   "test_metrics": {
-    "dice": 0.9234,
-    "iou": 0.8577,
-    "f1": 0.9234,
-    "precision": 0.9474,
-    "recall": 0.9005,
-    "accuracy": 0.9745
+    "dice": 0.9233709573745728,
+    "iou": 0.8576500415802002,
+    "f1": 0.9233709475438584,
+    "precision": 0.9474478438948487,
+    "recall": 0.9004874267475651,
+    "accuracy": 0.9744630432128907,
+    "loss": 0.09141451492905617
   },
   "dataset": "andreribeiro87/kvasir-seg-augmented",
   "hpo": "Optuna MedianPruner, 60 trials (28 completed, 32 pruned), best eval_loss=0.0537"