Added utility to convert images to greyscale using luminance weights (better for the current CNN structure).

Author: kmock930

framework/data_utils.py

@@ -6,11 +6,72 @@
 from datasets import load_from_disk
 from sklearn.model_selection import train_test_split
 from torch.utils.data import DataLoader
+from PIL import Image
 
 from framework import utils
 from framework.datasets import CIFAR10Dataset
 
 
+def convert_to_grayscale(image: np.ndarray) -> np.ndarray:
+    """Convert RGB/RGBA image to grayscale.
+    
+    Args:
+        image: Image array with shape (H, W, C) where C is 3 (RGB) or 4 (RGBA)
+        
+    Returns:
+        Grayscale image with shape (H, W) 
+    """
+    if len(image.shape) == 2:
+        # Already grayscale
+        return image
+    elif len(image.shape) == 3:
+        if image.shape[2] == 1:
+            # Single channel, just squeeze
+            return image.squeeze(axis=2)
+        elif image.shape[2] == 3:
+            # RGB -> Grayscale using luminance weights
+            # Source: https://www.songho.ca/dsp/luminance/luminance.html
+            return np.dot(image[...,:3], [0.299, 0.587, 0.114])
+        elif image.shape[2] == 4:
+            # RGBA -> Grayscale (ignore alpha)
+            return np.dot(image[...,:3], [0.299, 0.587, 0.114])
+    
+    raise ValueError(f"Unsupported image shape: {image.shape}")
+
+
+def preprocess_images_to_grayscale(images: List[np.ndarray]) -> List[np.ndarray]:
+    """Convert a list of images to grayscale.
+    
+    Args:
+        images: List of image arrays
+        
+    Returns:
+        List of grayscale image arrays
+    """
+    return [convert_to_grayscale(img) for img in images]
+
+
+def convert_dataset_to_grayscale(dataset):
+    """Convert HuggingFace dataset images to grayscale in-place preprocessing.
+    
+    Args:
+        dataset: HuggingFace dataset with 'image' column
+        
+    Returns:
+        List of grayscale images and labels
+    """
+    images = []
+    labels = []
+    
+    for item in dataset:
+        img = np.array(item['image'])
+        gray_img = convert_to_grayscale(img)
+        images.append(gray_img)
+        labels.append(item['label'])
+    
+    return images, np.array(labels)
+
+
 def load_cifar10_data():
     """Load CIFAR-10 dataset (grayscale from processed datasets)."""
     repo_root = Path(__file__).resolve().parents[1]