Update README.md

README.md

@@ -147,7 +147,7 @@ The model was trained on the **RVL-CDIP (Ryerson Vision Lab Complex Document Inf
 
 * **Total Size:** 400,000 grayscale images.
 * **Classes:** 16 perfectly balanced classes.
-* **Split:** The standard split is 320k Train, 40k Validation, 40k Test. This model was trained on the 25,000 images per class available in the training set.
+* **Split:** The standard split is 320k Train, 40k Validation, 40k Test. This model was trained on the 20k images per class with 2.5k Images per class for Val and 2.5k Images per class for Test.
 * **Data Handling:** Original grayscale images were converted to 3-channel RGB to match the input expectations of the pre-trained ResNet backbone.
 
 ### Training Procedure