Update README - Run 20251012_065325

README.md

@@ -21,7 +21,7 @@ model-index:
       type: imagenet-1k
     metrics:
     - type: accuracy
-      value: 82.94
+      value: 83.06
 ---
 
 # David: Multi-Scale Crystal Classifier
@@ -51,16 +51,17 @@ as class prototypes with role-weighted similarity computation (Rose Loss).
 ## Performance
 
 ### Best Results
-- **Validation Accuracy**: 82.94%
-- **Best Epoch**: 4
-- **Final Train Accuracy**: 89.32%
+- **Validation Accuracy**: 83.06%
+- **Best Epoch**: 5
+- **Final Train Accuracy**: 90.69%
 
 ### Per-Scale Performance
-- **Scale 256**: 82.94%
-- **Scale 512**: 82.93%
-- **Scale 768**: 82.86%
-- **Scale 1024**: 82.64%
-- **Scale 1280**: 82.31%
+- **Scale 256**: 83.06%
+- **Scale 512**: 83.06%
+- **Scale 768**: 83.00%
+- **Scale 1024**: 82.90%
+- **Scale 1280**: 82.74%
+- **Scale 1536**: 82.49%
 
 
 ## Usage
@@ -194,4 +195,4 @@ Special thanks to Claude (Anthropic) for debugging assistance.
 
 ---
 
-*Generated on 2025-10-12 07:48:28*
+*Generated on 2025-10-12 07:51:02*