create_dataset.py

import os
import shutil
import random
from tqdm import tqdm

def create_train_val_split(source_dir, output_dir, val_split=0.2):
    """
    Creates a train/val split from the PlantVillage dataset.
    
    Args:
        source_dir: Directory containing the PlantVillage dataset
        output_dir: Directory to save the split dataset
        val_split: Proportion of data to use for validation
    """
    # Create output directories
    train_dir = os.path.join(output_dir, 'train')
    val_dir = os.path.join(output_dir, 'val')
    
    os.makedirs(train_dir, exist_ok=True)
    os.makedirs(val_dir, exist_ok=True)
    
    # Get all class directories
    class_dirs = [d for d in os.listdir(source_dir) if os.path.isdir(os.path.join(source_dir, d))]
    
    for class_dir in tqdm(class_dirs, desc="Processing classes"):
        # Create class directories in train and val
        os.makedirs(os.path.join(train_dir, class_dir), exist_ok=True)
        os.makedirs(os.path.join(val_dir, class_dir), exist_ok=True)
        
        # Get all images in the class
        images = [f for f in os.listdir(os.path.join(source_dir, class_dir)) 
                 if f.endswith('.jpg') or f.endswith('.JPG') or f.endswith('.png')]
        
        # Shuffle images
        random.shuffle(images)
        
        # Split into train and val
        split_idx = int(len(images) * (1 - val_split))
        train_images = images[:split_idx]
        val_images = images[split_idx:]
        
        # Copy images to train directory
        for img in tqdm(train_images, desc=f"Copying {class_dir} train images", leave=False):
            src = os.path.join(source_dir, class_dir, img)
            dst = os.path.join(train_dir, class_dir, img)
            shutil.copy(src, dst)
        
        # Copy images to val directory
        for img in tqdm(val_images, desc=f"Copying {class_dir} val images", leave=False):
            src = os.path.join(source_dir, class_dir, img)
            dst = os.path.join(val_dir, class_dir, img)
            shutil.copy(src, dst)
    
    print(f"Dataset split complete. Train: {len(os.listdir(train_dir))} classes, Val: {len(os.listdir(val_dir))} classes")

def main():
    # Set paths
    source_dir = 'PlantVillage-Dataset/raw/color'  # Update this to your dataset path
    output_dir = 'PlantVillage'
    
    # Create the split
    create_train_val_split(source_dir, output_dir)

if __name__ == "__main__":
    main()