src/modules/cnn_deblurring.py

"""

CNN Deblurring Module - Deep Learning Based Image Enhancement

============================================================



CNN inference system for image deblurring with TensorFlow/Keras.

Includes model architecture, training utilities, and inference pipeline.

"""

import cv2
import numpy as np
import os
import logging
from typing import Optional, Tuple, List
import pickle

# Configure TensorFlow to reduce verbosity
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'  # Reduce TensorFlow logging
os.environ['TF_ENABLE_ONEDNN_OPTS'] = '0'  # Disable oneDNN messages

import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers, Model

# Configure TensorFlow settings
tf.get_logger().setLevel('ERROR')  # Only show errors
tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.ERROR)

# Configure logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

class CNNDeblurModel:
    """CNN-based deblurring model with encoder-decoder architecture"""
    
    def __init__(self, input_shape: Tuple[int, int, int] = (256, 256, 3)):
        self.input_shape = input_shape
        self.model = None
        self.is_trained = False
        self.training_history = None
        self.model_path = "models/cnn_deblur_model.h5"
        self.dataset_path = "data/training_dataset"
        
    def build_model(self) -> Model:
        """

        Build CNN deblurring model with U-Net like architecture

        

        Returns:

            keras.Model: Compiled CNN model

        """
        try:
            # Input layer
            inputs = keras.Input(shape=self.input_shape)
            
            # Encoder (Downsampling)
            conv1 = layers.Conv2D(64, 3, activation='relu', padding='same')(inputs)
            conv1 = layers.Conv2D(64, 3, activation='relu', padding='same')(conv1)
            pool1 = layers.MaxPooling2D(pool_size=(2, 2))(conv1)
            
            conv2 = layers.Conv2D(128, 3, activation='relu', padding='same')(pool1)
            conv2 = layers.Conv2D(128, 3, activation='relu', padding='same')(conv2)
            pool2 = layers.MaxPooling2D(pool_size=(2, 2))(conv2)
            
            conv3 = layers.Conv2D(256, 3, activation='relu', padding='same')(pool2)
            conv3 = layers.Conv2D(256, 3, activation='relu', padding='same')(conv3)
            pool3 = layers.MaxPooling2D(pool_size=(2, 2))(conv3)
            
            # Bottleneck
            conv4 = layers.Conv2D(512, 3, activation='relu', padding='same')(pool3)
            conv4 = layers.Conv2D(512, 3, activation='relu', padding='same')(conv4)
            
            # Decoder (Upsampling)
            up5 = layers.UpSampling2D(size=(2, 2))(conv4)
            up5 = layers.Conv2D(256, 2, activation='relu', padding='same')(up5)
            merge5 = layers.concatenate([conv3, up5], axis=3)
            conv5 = layers.Conv2D(256, 3, activation='relu', padding='same')(merge5)
            conv5 = layers.Conv2D(256, 3, activation='relu', padding='same')(conv5)
            
            up6 = layers.UpSampling2D(size=(2, 2))(conv5)
            up6 = layers.Conv2D(128, 2, activation='relu', padding='same')(up6)
            merge6 = layers.concatenate([conv2, up6], axis=3)
            conv6 = layers.Conv2D(128, 3, activation='relu', padding='same')(merge6)
            conv6 = layers.Conv2D(128, 3, activation='relu', padding='same')(conv6)
            
            up7 = layers.UpSampling2D(size=(2, 2))(conv6)
            up7 = layers.Conv2D(64, 2, activation='relu', padding='same')(up7)
            merge7 = layers.concatenate([conv1, up7], axis=3)
            conv7 = layers.Conv2D(64, 3, activation='relu', padding='same')(merge7)
            conv7 = layers.Conv2D(64, 3, activation='relu', padding='same')(conv7)
            
            # Output layer
            outputs = layers.Conv2D(3, 1, activation='sigmoid')(conv7)
            
            # Create model
            model = Model(inputs=inputs, outputs=outputs)
            
            # Compile model
            model.compile(
                optimizer='adam',
                loss='mse',
                metrics=['mae', 'mse']
            )
            
            self.model = model
            logger.info("CNN model built successfully")
            return model
            
        except Exception as e:
            logger.error(f"Error building CNN model: {e}")
            return None
    
    def load_model(self, model_path: str) -> bool:
        """

        Load pre-trained model from file

        

        Args:

            model_path: Path to saved model

        

        Returns:

            bool: Success status

        """
        try:
            if os.path.exists(model_path):
                self.model = keras.models.load_model(model_path)
                self.is_trained = True
                logger.info(f"Model loaded from {model_path}")
                return True
            else:
                logger.warning(f"Model file not found: {model_path}")
                # Build new model as fallback
                self.build_model()
                return False
                
        except Exception as e:
            logger.error(f"Error loading model: {e}")
            self.build_model()  # Fallback to new model
            return False
    
    def save_model(self, model_path: str) -> bool:
        """

        Save current model to file

        

        Args:

            model_path: Path to save model

        

        Returns:

            bool: Success status

        """
        try:
            if self.model is not None:
                self.model.save(model_path)
                logger.info(f"Model saved to {model_path}")
                return True
            else:
                logger.error("No model to save")
                return False
                
        except Exception as e:
            logger.error(f"Error saving model: {e}")
            return False
    
    def preprocess_image(self, image: np.ndarray) -> np.ndarray:
        """

        Preprocess image for CNN input with color preservation

        

        Args:

            image: Input image (BGR format)

        

        Returns:

            np.ndarray: Preprocessed image

        """
        try:
            # Convert BGR to RGB (preserve original precision)
            if len(image.shape) == 3 and image.shape[2] == 3:
                rgb_image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
            else:
                rgb_image = image
            
            # Resize to model input size with high-quality interpolation
            resized = cv2.resize(rgb_image, 
                               (self.input_shape[1], self.input_shape[0]), 
                               interpolation=cv2.INTER_CUBIC)  # Better color preservation
            
            # Normalize to [0, 1] with high precision
            normalized = resized.astype(np.float64) / 255.0  # Use float64 for precision
            
            # Add batch dimension
            batched = np.expand_dims(normalized, axis=0)
            
            return batched.astype(np.float32)  # Convert to float32 for model
            
        except Exception as e:
            logger.error(f"Error preprocessing image: {e}")
            return np.array([])
    
    def postprocess_image(self, output: np.ndarray, original_shape: Tuple[int, int]) -> np.ndarray:
        """

        Postprocess CNN output to original image format with color preservation

        

        Args:

            output: CNN model output

            original_shape: Original image shape (height, width)

        

        Returns:

            np.ndarray: Postprocessed image in BGR format

        """
        try:
            # Remove batch dimension
            if len(output.shape) == 4:
                output = output[0]
            
            # Denormalize from [0, 1] to [0, 255] with high precision
            denormalized = np.clip(output * 255.0, 0, 255)  # Clip before conversion
            denormalized = np.round(denormalized).astype(np.uint8)  # Round to preserve colors
            
            # Resize to original size with high-quality interpolation
            resized = cv2.resize(denormalized, 
                               (original_shape[1], original_shape[0]), 
                               interpolation=cv2.INTER_CUBIC)  # Better color preservation
            
            # Convert RGB back to BGR
            bgr_image = cv2.cvtColor(resized, cv2.COLOR_RGB2BGR)
            
            return bgr_image
            
        except Exception as e:
            logger.error(f"Error postprocessing image: {e}")
            return np.zeros((*original_shape, 3), dtype=np.uint8)
    
    def enhance_image(self, image: np.ndarray) -> np.ndarray:
        """

        Enhance image using CNN model

        

        Args:

            image: Input blurry image (BGR format)

        

        Returns:

            np.ndarray: Enhanced image (BGR format)

        """
        try:
            if self.model is None:
                logger.warning("No model available, building new model")
                self.build_model()
            
            # Store original shape
            original_shape = image.shape[:2]
            
            # Preprocess
            preprocessed = self.preprocess_image(image)
            
            if preprocessed.size == 0:
                logger.error("Failed to preprocess image")
                return image
            
            # If model is not trained, return enhanced version using traditional methods
            if not self.is_trained:
                logger.info("Using fallback enhancement (model not trained)")
                return self._fallback_enhancement(image)
            
            # CNN inference
            enhanced = self.model.predict(preprocessed, verbose=0)
            
            # Postprocess
            result = self.postprocess_image(enhanced, original_shape)
            
            logger.info("CNN enhancement completed")
            return result
            
        except Exception as e:
            logger.error(f"Error in CNN enhancement: {e}")
            return self._fallback_enhancement(image)
    
    def _fallback_enhancement(self, image: np.ndarray) -> np.ndarray:
        """

        Fallback enhancement when CNN model is not available - preserves original colors

        

        Args:

            image: Input image

        

        Returns:

            np.ndarray: Enhanced image using color-preserving traditional methods

        """
        try:
            # Method 1: Gentle unsharp masking with color preservation
            # Create a subtle blur for unsharp masking
            gaussian = cv2.GaussianBlur(image, (5, 5), 1.0)
            
            # Apply very gentle unsharp masking to avoid color shifts
            enhanced = cv2.addWeighted(image, 1.2, gaussian, -0.2, 0)
            
            # Method 2: Enhance sharpness without changing colors
            # Convert to float for precision
            img_float = image.astype(np.float64)
            
            # Apply high-pass filter for sharpening
            kernel_sharpen = np.array([[-0.1, -0.1, -0.1],
                                     [-0.1,  1.8, -0.1], 
                                     [-0.1, -0.1, -0.1]])
            
            # Apply sharpening kernel to each channel separately
            sharpened_channels = []
            for i in range(3):  # Process each color channel
                channel = img_float[:, :, i]
                sharpened_channel = cv2.filter2D(channel, -1, kernel_sharpen)
                sharpened_channels.append(sharpened_channel)
            
            sharpened = np.stack(sharpened_channels, axis=2)
            
            # Combine original with sharpened (gentle blend)
            result = 0.7 * img_float + 0.3 * sharpened
            
            # Carefully clip and convert back
            result = np.clip(result, 0, 255).astype(np.uint8)
            
            logger.info("Color-preserving fallback enhancement applied")
            return result
            
        except Exception as e:
            logger.error(f"Error in fallback enhancement: {e}")
            return image

class CNNTrainer:
    """Training utilities for CNN deblurring model"""
    
    def __init__(self, model: CNNDeblurModel):
        self.model = model
    
    def create_synthetic_data(self, clean_images: List[np.ndarray], 

                            blur_types: List[str] = None) -> Tuple[np.ndarray, np.ndarray]:
        """

        Create synthetic training data by applying blur to clean images

        

        Args:

            clean_images: List of clean images

            blur_types: Types of blur to apply

        

        Returns:

            tuple: (blurred_images, clean_images) for training

        """
        if blur_types is None:
            blur_types = ['gaussian', 'motion', 'defocus']
        
        blurred_batch = []
        clean_batch = []
        
        try:
            for clean_img in clean_images:
                # Random blur type
                blur_type = np.random.choice(blur_types)
                
                if blur_type == 'gaussian':
                    # Gaussian blur
                    kernel_size = np.random.randint(5, 15)
                    if kernel_size % 2 == 0:
                        kernel_size += 1
                    blurred = cv2.GaussianBlur(clean_img, (kernel_size, kernel_size), 0)
                
                elif blur_type == 'motion':
                    # Motion blur
                    length = np.random.randint(5, 20)
                    angle = np.random.randint(0, 180)
                    kernel = self._create_motion_kernel(length, angle)
                    blurred = cv2.filter2D(clean_img, -1, kernel)
                
                else:  # defocus
                    # Defocus blur (approximated with Gaussian)
                    sigma = np.random.uniform(1, 5)
                    blurred = cv2.GaussianBlur(clean_img, (0, 0), sigma)
                
                blurred_batch.append(blurred)
                clean_batch.append(clean_img)
            
            return np.array(blurred_batch), np.array(clean_batch)
            
        except Exception as e:
            logger.error(f"Error creating synthetic data: {e}")
            return np.array([]), np.array([])
    
    def _create_motion_kernel(self, length: int, angle: float) -> np.ndarray:
        """Create motion blur kernel"""
        kernel = np.zeros((length, length))
        center = length // 2
        
        cos_val = np.cos(np.radians(angle))
        sin_val = np.sin(np.radians(angle))
        
        for i in range(length):
            offset = i - center
            y = int(center + offset * sin_val)
            x = int(center + offset * cos_val)
            if 0 <= y < length and 0 <= x < length:
                kernel[y, x] = 1
        
        return kernel / kernel.sum()
    
    def _load_user_images(self) -> List[np.ndarray]:
        """Load user's training images from training_dataset folder"""
        user_images = []
        
        try:
            if not os.path.exists(self.dataset_path):
                return user_images
                
            # Supported image extensions
            valid_extensions = {'.jpg', '.jpeg', '.png', '.bmp', '.tiff', '.tif'}
            
            for filename in os.listdir(self.dataset_path):
                if any(filename.lower().endswith(ext) for ext in valid_extensions):
                    image_path = os.path.join(self.dataset_path, filename)
                    try:
                        # Load image
                        image = cv2.imread(image_path)
                        if image is not None:
                            # Resize to model input size
                            resized = cv2.resize(image, (self.input_shape[1], self.input_shape[0]))
                            user_images.append(resized)
                            logger.info(f"Loaded user image: {filename}")
                    except Exception as e:
                        logger.warning(f"Failed to load {filename}: {e}")
            
            logger.info(f"Loaded {len(user_images)} user training images")
            return user_images
            
        except Exception as e:
            logger.error(f"Error loading user images: {e}")
            return []

    def create_training_dataset(self, num_samples: int = 1000, save_dataset: bool = True) -> Tuple[np.ndarray, np.ndarray]:
        """

        Create comprehensive training dataset with various blur types

        Incorporates user's real training images from data/training_dataset/

        

        Args:

            num_samples: Number of training samples to generate

            save_dataset: Whether to save dataset to disk

        

        Returns:

            Tuple[np.ndarray, np.ndarray]: Blurred images and clean targets

        """
        try:
            logger.info(f"Creating training dataset with {num_samples} samples...")
            
            # Ensure dataset directory exists
            os.makedirs(self.dataset_path, exist_ok=True)
            
            # Load user's training images
            user_images = self._load_user_images()
            
            all_blurred = []
            all_clean = []
            
            # First, process user images if available
            if user_images:
                logger.info(f"Processing {len(user_images)} user training images...")
                for user_img in user_images:
                    # Use user image as clean target multiple times with different blur types
                    for _ in range(3):  # Create 3 variations per user image
                        blur_type = np.random.choice(['gaussian', 'motion', 'defocus'])
                        
                        if blur_type == 'gaussian':
                            sigma = np.random.uniform(0.5, 3.0)
                            blurred = cv2.GaussianBlur(user_img, (0, 0), sigma)
                        elif blur_type == 'motion':
                            length = np.random.randint(5, 25)
                            angle = np.random.randint(0, 180)
                            kernel = self._create_motion_kernel(length, angle)
                            blurred = cv2.filter2D(user_img, -1, kernel)
                        else:  # defocus
                            sigma = np.random.uniform(1.0, 4.0)
                            blurred = cv2.GaussianBlur(user_img, (0, 0), sigma)
                        
                        # Add slight noise for realism
                        noise = np.random.normal(0, 3, blurred.shape).astype(np.float32)
                        blurred = np.clip(blurred.astype(np.float32) + noise, 0, 255).astype(np.uint8)
                        
                        all_blurred.append(blurred)
                        all_clean.append(user_img)
            
            # Generate remaining samples with synthetic images
            remaining_samples = max(0, num_samples - len(all_blurred))
            if remaining_samples > 0:
                logger.info(f"Generating {remaining_samples} synthetic training samples...")
                
                batch_size = 50
                num_batches = (remaining_samples + batch_size - 1) // batch_size
                
                for batch_idx in range(num_batches):
                    current_batch_size = min(batch_size, remaining_samples - batch_idx * batch_size)
                    
                    # Create synthetic clean images
                    clean_batch = self._generate_clean_images(current_batch_size)
                
                # Apply various blur types
                blurred_batch = []
                for clean_img in clean_batch:
                    blur_type = np.random.choice(['gaussian', 'motion', 'defocus'])
                    
                    if blur_type == 'gaussian':
                        sigma = np.random.uniform(0.5, 3.0)
                        blurred = cv2.GaussianBlur(clean_img, (0, 0), sigma)
                    elif blur_type == 'motion':
                        length = np.random.randint(5, 25)
                        angle = np.random.randint(0, 180)
                        kernel = self._create_motion_kernel(length, angle)
                        blurred = cv2.filter2D(clean_img, -1, kernel)
                    else:  # defocus
                        sigma = np.random.uniform(1.0, 4.0)
                        blurred = cv2.GaussianBlur(clean_img, (0, 0), sigma)
                    
                    # Add slight noise for realism
                    noise = np.random.normal(0, 5, blurred.shape).astype(np.float32)
                    blurred = np.clip(blurred.astype(np.float32) + noise, 0, 255).astype(np.uint8)
                    
                    blurred_batch.append(blurred)
                
                all_blurred.extend(blurred_batch)
                all_clean.extend(clean_batch)
                
                if (batch_idx + 1) % 5 == 0:
                    logger.info(f"Generated batch {batch_idx + 1}/{num_batches}")
            
            # Convert to numpy arrays
            blurred_dataset = np.array(all_blurred)
            clean_dataset = np.array(all_clean)
            
            # Normalize to [0, 1]
            blurred_dataset = blurred_dataset.astype(np.float32) / 255.0
            clean_dataset = clean_dataset.astype(np.float32) / 255.0
            
            logger.info(f"Dataset created: {blurred_dataset.shape} blurred, {clean_dataset.shape} clean")
            
            # Save dataset if requested
            if save_dataset:
                np.save(os.path.join(self.dataset_path, 'blurred_images.npy'), blurred_dataset)
                np.save(os.path.join(self.dataset_path, 'clean_images.npy'), clean_dataset)
                logger.info(f"Dataset saved to {self.dataset_path}")
            
            return blurred_dataset, clean_dataset
            
        except Exception as e:
            logger.error(f"Error creating training dataset: {e}")
            return np.array([]), np.array([])
    
    def _generate_clean_images(self, num_images: int) -> List[np.ndarray]:
        """Generate synthetic clean images for training"""
        clean_images = []
        
        for _ in range(num_images):
            # Create random patterns and shapes
            img = np.zeros((self.input_shape[0], self.input_shape[1], 3), dtype=np.uint8)
            
            # Random background
            bg_color = np.random.randint(0, 255, 3)
            img[:] = bg_color
            
            # Add random shapes
            num_shapes = np.random.randint(3, 8)
            for _ in range(num_shapes):
                shape_type = np.random.choice(['rectangle', 'circle', 'line'])
                color = np.random.randint(0, 255, 3).tolist()
                
                if shape_type == 'rectangle':
                    pt1 = (np.random.randint(0, img.shape[1]//2), np.random.randint(0, img.shape[0]//2))
                    pt2 = (np.random.randint(img.shape[1]//2, img.shape[1]), 
                          np.random.randint(img.shape[0]//2, img.shape[0]))
                    cv2.rectangle(img, pt1, pt2, color, -1)
                    
                elif shape_type == 'circle':
                    center = (np.random.randint(0, img.shape[1]), np.random.randint(0, img.shape[0]))
                    radius = np.random.randint(10, 50)
                    cv2.circle(img, center, radius, color, -1)
                    
                else:  # line
                    pt1 = (np.random.randint(0, img.shape[1]), np.random.randint(0, img.shape[0]))
                    pt2 = (np.random.randint(0, img.shape[1]), np.random.randint(0, img.shape[0]))
                    thickness = np.random.randint(1, 5)
                    cv2.line(img, pt1, pt2, color, thickness)
            
            # Add random text
            if np.random.random() > 0.5:
                text = ''.join(np.random.choice(list('ABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789'), 
                                              np.random.randint(3, 8)))
                font = cv2.FONT_HERSHEY_SIMPLEX
                font_scale = np.random.uniform(0.5, 2.0)
                color = np.random.randint(0, 255, 3).tolist()
                thickness = np.random.randint(1, 3)
                position = (np.random.randint(0, img.shape[1]//2), np.random.randint(20, img.shape[0]))
                cv2.putText(img, text, position, font, font_scale, color, thickness)
            
            clean_images.append(img)
        
        return clean_images
    
    def load_existing_dataset(self) -> Tuple[Optional[np.ndarray], Optional[np.ndarray]]:
        """Load existing dataset from disk"""
        try:
            blurred_path = os.path.join(self.dataset_path, 'blurred_images.npy')
            clean_path = os.path.join(self.dataset_path, 'clean_images.npy')
            
            if os.path.exists(blurred_path) and os.path.exists(clean_path):
                blurred_data = np.load(blurred_path)
                clean_data = np.load(clean_path)
                logger.info(f"Loaded existing dataset: {blurred_data.shape} samples")
                return blurred_data, clean_data
            else:
                logger.info("No existing dataset found")
                return None, None
                
        except Exception as e:
            logger.error(f"Error loading existing dataset: {e}")
            return None, None
    
    def train_model(self, 

                   epochs: int = 20, 

                   batch_size: int = 16, 

                   validation_split: float = 0.2,

                   use_existing_dataset: bool = True,

                   num_training_samples: int = 1000) -> bool:
        """

        Train the CNN model with comprehensive dataset

        

        Args:

            epochs: Number of training epochs

            batch_size: Training batch size

            validation_split: Fraction of data for validation

            use_existing_dataset: Whether to use existing saved dataset

            num_training_samples: Number of samples to generate if creating new dataset

        

        Returns:

            bool: Training success status

        """
        try:
            logger.info("Starting CNN model training...")
            
            # Build model if not exists
            if self.model is None:
                self.build_model()
            
            # Load or create dataset
            if use_existing_dataset:
                blurred_data, clean_data = self.load_existing_dataset()
                if blurred_data is None:
                    logger.info("Creating new dataset...")
                    blurred_data, clean_data = self.create_training_dataset(num_training_samples)
            else:
                logger.info("Creating new dataset...")
                blurred_data, clean_data = self.create_training_dataset(num_training_samples)
            
            if len(blurred_data) == 0:
                logger.error("Failed to create/load training dataset")
                return False
            
            logger.info(f"Training on {len(blurred_data)} samples")
            
            # Setup callbacks
            callbacks = [
                keras.callbacks.EarlyStopping(
                    monitor='val_loss', 
                    patience=5, 
                    restore_best_weights=True
                ),
                keras.callbacks.ReduceLROnPlateau(
                    monitor='val_loss', 
                    factor=0.5, 
                    patience=3, 
                    min_lr=1e-7
                ),
                keras.callbacks.ModelCheckpoint(
                    filepath=self.model_path,
                    monitor='val_loss',
                    save_best_only=True,
                    save_weights_only=False
                )
            ]
            
            # Train model
            self.training_history = self.model.fit(
                blurred_data, clean_data,
                epochs=epochs,
                batch_size=batch_size,
                validation_split=validation_split,
                callbacks=callbacks,
                verbose=1
            )
            
            # Save final model
            self.save_model(self.model_path)
            self.is_trained = True
            
            # Save training history
            history_path = self.model_path.replace('.h5', '_history.pkl')
            with open(history_path, 'wb') as f:
                pickle.dump(self.training_history.history, f)
            
            logger.info("Training completed successfully!")
            logger.info(f"Model saved to: {self.model_path}")
            
            # Print training summary
            final_loss = self.training_history.history['loss'][-1]
            final_val_loss = self.training_history.history['val_loss'][-1]
            logger.info(f"Final training loss: {final_loss:.4f}")
            logger.info(f"Final validation loss: {final_val_loss:.4f}")
            
            return True
            
        except Exception as e:
            logger.error(f"Error during training: {e}")
            return False
    
    def evaluate_model(self, test_images: np.ndarray = None, test_targets: np.ndarray = None) -> dict:
        """

        Evaluate model performance on test data

        

        Args:

            test_images: Test images (if None, creates synthetic test set)

            test_targets: Test targets (if None, creates synthetic test set)

        

        Returns:

            dict: Evaluation metrics

        """
        try:
            if self.model is None or not self.is_trained:
                logger.error("Model not trained. Train the model first.")
                return {}
            
            # Create test data if not provided
            if test_images is None or test_targets is None:
                logger.info("Creating test dataset...")
                test_images, test_targets = self.create_training_dataset(num_samples=100, save_dataset=False)
            
            # Evaluate
            results = self.model.evaluate(test_images, test_targets, verbose=0)
            
            metrics = {
                'loss': results[0],
                'mae': results[1],
                'mse': results[2]
            }
            
            logger.info("Model Evaluation Results:")
            for metric, value in metrics.items():
                logger.info(f"  {metric}: {value:.4f}")
            
            return metrics
            
        except Exception as e:
            logger.error(f"Error during evaluation: {e}")
            return {}

# Convenience functions
def load_cnn_model(model_path: str = "models/cnn_model.h5") -> CNNDeblurModel:
    """

    Load CNN deblurring model

    

    Args:

        model_path: Path to model file

    

    Returns:

        CNNDeblurModel: Loaded model instance

    """
    model = CNNDeblurModel()
    model.load_model(model_path)
    return model

def enhance_with_cnn(image: np.ndarray, model_path: str = "models/cnn_model.h5") -> np.ndarray:
    """

    Enhance image using CNN model

    

    Args:

        image: Input image

        model_path: Path to model file

    

    Returns:

        np.ndarray: Enhanced image

    """
    model = load_cnn_model(model_path)
    return model.enhance_image(image)

# Training utility functions
def train_new_model(num_samples: int = 1000, epochs: int = 20, input_shape: Tuple[int, int, int] = (256, 256, 3)):
    """

    Train a new CNN deblurring model from scratch

    

    Args:

        num_samples: Number of training samples to generate

        epochs: Number of training epochs

        input_shape: Input image shape

    

    Returns:

        CNNDeblurModel: Trained model

    """
    print("🚀 Training New CNN Deblurring Model")
    print("=" * 50)
    
    # Ensure directories exist
    os.makedirs("models", exist_ok=True)
    os.makedirs("data/training_dataset", exist_ok=True)
    
    # Initialize model
    model = CNNDeblurModel(input_shape=input_shape)
    
    # Train model
    success = model.train_model(
        epochs=epochs,
        batch_size=16,
        validation_split=0.2,
        use_existing_dataset=True,
        num_training_samples=num_samples
    )
    
    if success:
        print("✅ Training completed successfully!")
        
        # Evaluate model
        metrics = model.evaluate_model()
        if metrics:
            print(f"📊 Model Performance:")
            print(f"   Loss: {metrics['loss']:.4f}")
            print(f"   MAE: {metrics['mae']:.4f}")
            print(f"   MSE: {metrics['mse']:.4f}")
        
        return model
    else:
        print("❌ Training failed!")
        return None

def quick_train():
    """Quick training with default parameters"""
    return train_new_model(num_samples=500, epochs=10)

def full_train():
    """Full training with comprehensive dataset"""
    return train_new_model(num_samples=2000, epochs=30)

# Example usage and testing
if __name__ == "__main__":
    import argparse
    
    parser = argparse.ArgumentParser(description='CNN Deblurring Module')
    parser.add_argument('--train', action='store_true', help='Train the model')
    parser.add_argument('--quick-train', action='store_true', help='Quick training (500 samples, 10 epochs)')
    parser.add_argument('--full-train', action='store_true', help='Full training (2000 samples, 30 epochs)')
    parser.add_argument('--samples', type=int, default=1000, help='Number of training samples')
    parser.add_argument('--epochs', type=int, default=20, help='Number of training epochs')
    parser.add_argument('--test', action='store_true', help='Test the model')
    
    args = parser.parse_args()
    
    print("🎯 CNN Deblurring Module")
    print("=" * 30)
    
    if args.quick_train:
        print("🚀 Quick Training Mode")
        model = quick_train()
        
    elif args.full_train:
        print("🚀 Full Training Mode")
        model = full_train()
        
    elif args.train:
        print(f"🚀 Custom Training Mode")
        model = train_new_model(num_samples=args.samples, epochs=args.epochs)
        
    elif args.test:
        print("🧪 Testing Mode")
        
        # Create test image
        test_image = np.random.randint(0, 255, (480, 640, 3), dtype=np.uint8)
        
        # Initialize model
        cnn_model = CNNDeblurModel()
        
        # Try to load existing model
        if cnn_model.load_model(cnn_model.model_path):
            print(f"✅ Loaded existing trained model")
        else:
            print(f"ℹ️ No trained model found, building new model")
            cnn_model.build_model()
        
        print(f"Model input shape: {cnn_model.input_shape}")
        print(f"Model built: {cnn_model.model is not None}")
        print(f"Model trained: {cnn_model.is_trained}")
        
        # Test enhancement
        enhanced = cnn_model.enhance_image(test_image)
        print(f"Original shape: {test_image.shape}")
        print(f"Enhanced shape: {enhanced.shape}")
        
        if cnn_model.is_trained:
            # Evaluate on test data
            metrics = cnn_model.evaluate_model()
            if metrics:
                print("📊 Model Performance:")
                for metric, value in metrics.items():
                    print(f"   {metric}: {value:.4f}")
    
    else:
        print("ℹ️ Usage options:")
        print("  --test          Test existing model or build new one")
        print("  --quick-train   Quick training (500 samples, 10 epochs)")
        print("  --full-train    Full training (2000 samples, 30 epochs)")
        print("  --train         Custom training (use --samples and --epochs)")
        print("\nExamples:")
        print("  python -m modules.cnn_deblurring --test")
        print("  python -m modules.cnn_deblurring --quick-train")
        print("  python -m modules.cnn_deblurring --train --samples 1500 --epochs 25")
    
    print("\n🎯 CNN deblurring module ready!")