inference.py

import os
import joblib
import numpy as np
import tensorflow as tf
from keras.models import load_model
import cv2
import matplotlib.pyplot as plt
from PIL import Image
import io


class FakeImageDetector:
    def __init__(self, feature_extractor_path="hybrid_model_weights.h5", 
                 classifier_path="gbdt_model.pkl", img_size=224):
        self.img_size = img_size
        self.feature_extractor = None
        self.classifier = None
        self.load_models(feature_extractor_path, classifier_path)
    
    def load_models(self, feature_extractor_path, classifier_path):
        """Load pre-trained models"""
        print("🔄 Loading models...")
        self.feature_extractor = load_model(feature_extractor_path, compile=False)
        self.classifier = joblib.load(classifier_path)
        print("✅ Models loaded successfully")
    
    def preprocess_image(self, image_input):
        """
        Preprocess image from various inputs
        Supports: file path, numpy array, PIL Image
        """
        try:
            # Handle different input types
            if isinstance(image_input, str):  # File path
                img = cv2.imread(image_input)
                if img is None:
                    raise ValueError(f"Could not read image from {image_input}")
            elif isinstance(image_input, np.ndarray):  # Numpy array
                img = image_input.copy()
                if len(img.shape) == 2:  # Grayscale
                    img = cv2.cvtColor(img, cv2.COLOR_GRAY2BGR)
                elif img.shape[2] == 4:  # RGBA
                    img = cv2.cvtColor(img, cv2.COLOR_RGBA2BGR)
            elif hasattr(image_input, 'read'):  # File-like object
                img_array = np.frombuffer(image_input.read(), np.uint8)
                img = cv2.imdecode(img_array, cv2.IMREAD_COLOR)
            else:  # Assume PIL Image
                img = cv2.cvtColor(np.array(image_input), cv2.COLOR_RGB2BGR)
            
            # Convert to RGB
            img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
            
            # Resize
            img = cv2.resize(img, (self.img_size, self.img_size), interpolation=cv2.INTER_AREA)
            img = img.astype('float32') / 255.0
            
            return img
        except Exception as e:
            print(f"Error preprocessing image: {str(e)}")
            return None
    
    def extract_features(self, image):
        """Extract features from preprocessed image"""
        image_batch = np.expand_dims(image, axis=0)
        features = self.feature_extractor.predict(image_batch, verbose=0)
        return features
    
    def predict(self, image_input, return_confidence=False):
        """
        Predict if image is real or fake
        
        Args:
            image_input: Can be file path, numpy array, or PIL Image
            return_confidence: If True, returns confidence score too
        
        Returns:
            prediction (0=Real, 1=Fake) and optional confidence
        """
        # Preprocess image
        processed_image = self.preprocess_image(image_input)
        if processed_image is None:
            return None
        
        # Extract features
        features = self.extract_features(processed_image)
        
        # Make prediction
        prediction = self.classifier.predict(features)[0]
        confidence = self.classifier.predict_proba(features)[0][prediction] * 100
        
        result = {
            'prediction': 'Real' if prediction == 0 else 'Fake',
            'confidence': confidence,
            'prediction_code': int(prediction)
        }
        
        if return_confidence:
            return result
        else:
            return result['prediction']
    
    def predict_batch(self, image_paths):
        """Predict batch of images"""
        results = []
        for img_path in image_paths:
            result = self.predict(img_path, return_confidence=True)
            if result:
                results.append({
                    'image_path': img_path,
                    'prediction': result['prediction'],
                    'confidence': result['confidence']
                })
        return results
    
    def visualize_prediction(self, image_input, save_path=None):
        """Make prediction and visualize result"""
        result = self.predict(image_input, return_confidence=True)
        if result is None:
            print("❌ Could not process image")
            return
        
        # Load image for display
        if isinstance(image_input, str):
            display_img = cv2.imread(image_input)
            display_img = cv2.cvtColor(display_img, cv2.COLOR_BGR2RGB)
        else:
            display_img = self.preprocess_image(image_input)
            display_img = (display_img * 255).astype(np.uint8)
        
        # Create visualization
        plt.figure(figsize=(8, 6))
        plt.imshow(display_img)
        plt.title(f"Prediction: {result['prediction']} ({result['confidence']:.1f}%)", 
                 fontsize=16, pad=20)
        
        # Color code based on prediction
        color = 'green' if result['prediction'] == 'Real' else 'red'
        plt.gca().text(0.5, -0.1, 
                      f"Confidence: {result['confidence']:.1f}%", 
                      ha='center', va='center', 
                      transform=plt.gca().transAxes,
                      fontsize=12,
                      bbox=dict(boxstyle="round,pad=0.3", 
                               facecolor=color, 
                               alpha=0.5))
        
        plt.axis('off')
        
        if save_path:
            plt.savefig(save_path, dpi=100, bbox_inches='tight')
            print(f"✅ Visualization saved to {save_path}")
        
        plt.show()
        return result


# Example usage
if __name__ == "__main__":
    # Initialize detector
    detector = FakeImageDetector()
    
    # Example 1: Predict from file path
    if os.path.exists("real_0.jpg"):
        result = detector.predict("fake_0.jpg", return_confidence=True)
        print(f"Prediction: {result['prediction']} ({result['confidence']:.1f}%)")
        
        # Visualize
        detector.visualize_prediction("fake_0.jpg")
    
    # Example 2: Predict from URL (requires additional libraries)
    # import requests
    # response = requests.get("https://example.com/image.jpg")
    # result = detector.predict(io.BytesIO(response.content))
    
    # Example 3: Batch prediction
    # image_list = ["img1.jpg", "img2.jpg", "img3.jpg"]
    # results = detector.predict_batch(image_list)
    # for res in results:
    #     print(f"{res['image_path']}: {res['prediction']} ({res['confidence']:.1f}%)")