Update README.md

README.md

@@ -69,6 +69,22 @@ AutoGluon automatically searched across the following dimensions:
   - Mixup (interpolated samples).  
 - **Early stopping:** Triggered automatically when validation metric stops improving.
 
+---
+### Augmentation Pipeline
+- Random resized crop to **224 × 224 pixels**  
+- Random horizontal flip  
+- Color jitter (brightness, contrast, saturation, hue)  
+- RandAugment (random transformations applied with strength parameter)  
+- Mixup with α = 0.2 (blending images/labels)  
+
+### Input Resolution
+- All images resized to **224 × 224** before being passed to the network  
+
+### Expected Preprocessing
+- RGB image normalization (mean/std) using ImageNet statistics  
+- One hot encoding of labels for classification  
+- Train/validation split: 80/20 stratified
+
 ---
 
 ## Results
@@ -108,6 +124,15 @@ On the held-out **original split** (~40 images):
 - Labeling reflects simplified categories; cultural/geographic nuance lost.  
 
 ---
+
+## Known Failure Modes
+- Struggles on images with unusual lighting/backgrounds  
+- Misclassifies foods with **fusion characteristics** (e.g., Asian inspired Western dishes)  
+- Sensitive to **out-of-distribution inputs** (images outside the dataset’s augmentation domain)  
+- Performs poorly when food is occluded or partially cropped
+
+---
+
 ## AI Usage Disclosure
 
 - Assistance tools were used to streamline coding