inference_example.py

#!/usr/bin/env python3
"""
Person Classification TensorFlow Lite Inference Example

This script demonstrates how to use the person classification TFLite models
for binary classification (person vs. non-person) on input images.

Usage:
    python inference_example.py --model flash --image test_image.jpg
    python inference_example.py --model sram --image test_image.jpg
"""

import argparse
import sys
import os
import time
import numpy as np
import tensorflow as tf
from PIL import Image
from pathlib import Path

class PersonClassifier:
    """Person Classification using TensorFlow Lite models"""
    
    def __init__(self, model_path):
        """
        Initialize the classifier with a TFLite model
        
        Args:
            model_path (str): Path to the .tflite model file
        """
        self.model_path = model_path
        self.interpreter = None
        self.input_details = None
        self.output_details = None
        self.input_shape = None
        self.load_model()
    
    def load_model(self):
        """Load the TensorFlow Lite model"""
        try:
            self.interpreter = tf.lite.Interpreter(model_path=self.model_path)
            self.interpreter.allocate_tensors()
            
            self.input_details = self.interpreter.get_input_details()
            self.output_details = self.interpreter.get_output_details()
            self.input_shape = self.input_details[0]['shape']
            
            print(f"✅ Model loaded successfully: {self.model_path}")
            print(f"📊 Input shape: {self.input_shape}")
            print(f"📊 Input dtype: {self.input_details[0]['dtype']}")
            print(f"📊 Output shape: {self.output_details[0]['shape']}")
            print(f"📊 Output dtype: {self.output_details[0]['dtype']}")
            
        except Exception as e:
            print(f"❌ Error loading model: {e}")
            sys.exit(1)
    
    def preprocess_image(self, image_path):
        """
        Preprocess input image for model inference
        
        Args:
            image_path (str): Path to input image
            
        Returns:
            np.ndarray: Preprocessed image array ready for inference
        """
        try:
            # Load and convert image
            image = Image.open(image_path).convert('RGB')
            print(f"📸 Original image size: {image.size}")
            
            # Get target size from model input shape (height, width)
            target_height = self.input_shape[1]
            target_width = self.input_shape[2]
            target_size = (target_width, target_height)  # PIL uses (width, height)
            
            # Resize image to model's expected input size
            image = image.resize(target_size, Image.Resampling.BILINEAR)
            print(f"🔄 Resized to: {target_size} (W×H)")
            
            # Convert to numpy array
            image_array = np.array(image, dtype=np.uint8)
            
            # Add batch dimension [batch, height, width, channels]
            image_batch = np.expand_dims(image_array, axis=0)
            
            print(f"✅ Preprocessed shape: {image_batch.shape}")
            print(f"📊 Value range: [{image_batch.min()}, {image_batch.max()}]")
            
            return image_batch
            
        except Exception as e:
            print(f"❌ Error preprocessing image: {e}")
            sys.exit(1)
    
    def predict(self, image_data):
        """
        Run inference on preprocessed image data
        
        Args:
            image_data (np.ndarray): Preprocessed image data
            
        Returns:
            tuple: (probability, prediction_label, confidence)
        """
        try:
            # Set input tensor
            self.interpreter.set_tensor(self.input_details[0]['index'], image_data)
            
            # Run inference
            start_time = time.time()
            self.interpreter.invoke()
            inference_time = time.time() - start_time
            
            # Get output tensor
            output_data = self.interpreter.get_tensor(self.output_details[0]['index'])
            
            # Handle quantized vs float output
            scale = self.output_details[0]['quantization'][0]
            zero_point = self.output_details[0]['quantization'][1]
            
            if scale != 0:  # Quantized output
                # Dequantize
                dequantized_output = scale * (output_data.astype(np.float32) - zero_point)
                # Apply sigmoid to get probability
                probability = 1 / (1 + np.exp(-dequantized_output[0][0]))
                print(f"🔢 Quantized output dequantized: {dequantized_output[0][0]:.4f}")
            else:  # Float output
                probability = float(output_data[0][0])
            
            # Determine prediction
            prediction_label = "Person" if probability > 0.5 else "Non-person"
            confidence = probability if probability > 0.5 else (1 - probability)
            
            print(f"⏱️  Inference time: {inference_time*1000:.2f}ms")
            
            return probability, prediction_label, confidence
            
        except Exception as e:
            print(f"❌ Error during inference: {e}")
            sys.exit(1)
    
    def classify_image(self, image_path):
        """
        Complete pipeline: preprocess image and run classification
        
        Args:
            image_path (str): Path to input image
            
        Returns:
            dict: Classification results
        """
        print(f"\n🔍 Classifying image: {image_path}")
        print("=" * 50)
        
        # Preprocess image
        image_data = self.preprocess_image(image_path)
        
        # Run inference
        probability, prediction_label, confidence = self.predict(image_data)
        
        # Compile results
        results = {
            'image_path': image_path,
            'prediction': prediction_label,
            'probability': probability,
            'confidence': confidence,
            'model_used': self.model_path
        }
        
        return results

def print_results(results):
    """Print classification results in a formatted way"""
    print("\n📋 CLASSIFICATION RESULTS")
    print("=" * 50)
    print(f"🖼️  Image: {results['image_path']}")
    print(f"🎯 Prediction: {results['prediction']}")
    print(f"📊 Probability: {results['probability']:.4f}")
    print(f"✅ Confidence: {results['confidence']:.1%}")
    print(f"🤖 Model: {Path(results['model_used']).name}")
    print("=" * 50)

def main():
    parser = argparse.ArgumentParser(
        description="Person Classification using TensorFlow Lite models",
        formatter_class=argparse.RawDescriptionHelpFormatter,
        epilog="""
Examples:
  python inference_example.py --model flash --image test_image.jpg
  python inference_example.py --model sram --image test_image.jpg
  python inference_example.py --model person_classification_flash(448x640).tflite --image test_image.jpg
        """
    )
    
    parser.add_argument(
        '--model', 
        required=True,
        help='Model to use: "flash", "sram", or path to .tflite file'
    )
    
    parser.add_argument(
        '--image',
        required=True,
        help='Path to input image file'
    )
    
    args = parser.parse_args()
    
    # Determine model path
    if args.model.lower() == 'flash':
        model_path = 'person_classification_flash(448x640).tflite'
    elif args.model.lower() == 'sram':
        model_path = 'person_classification_sram(256x448).tflite'
    else:
        model_path = args.model
    
    # Check if model file exists
    if not os.path.exists(model_path):
        print(f"❌ Model file not found: {model_path}")
        sys.exit(1)
    
    # Check if image file exists  
    if not os.path.exists(args.image):
        print(f"❌ Image file not found: {args.image}")
        sys.exit(1)
    
    print("🚀 Person Classification TensorFlow Lite Demo")
    print("=" * 50)
    
    # Initialize classifier
    classifier = PersonClassifier(model_path)
    
    # Run classification
    results = classifier.classify_image(args.image)
    
    # Print results
    print_results(results)

def demo_both_models():
    """Demo function to test both models if available"""
    print("🚀 Person Classification Demo - Both Models")
    print("=" * 50)
    
    models = [
        ('Flash Model (VGA)', 'person_classification_flash(448x640).tflite'),
        ('SRAM Model (WQVGA)', 'person_classification_sram(256x448).tflite')
    ]
    
    # Create a simple test image if none exists
    test_image_path = 'test_person.jpg'
    if not os.path.exists(test_image_path):
        print(f"ℹ️  Creating test image: {test_image_path}")
        # Create a simple test image (colored rectangle)
        test_img = Image.new('RGB', (640, 480), color='lightblue')
        test_img.save(test_image_path)
    
    for model_name, model_path in models:
        if os.path.exists(model_path):
            print(f"\n🔍 Testing {model_name}")
            print("-" * 30)
            
            classifier = PersonClassifier(model_path)
            results = classifier.classify_image(test_image_path)
            print_results(results)
        else:
            print(f"⚠️  {model_name} not found: {model_path}")

if __name__ == '__main__':
    if len(sys.argv) == 1:
        # If no arguments provided, run demo
        demo_both_models()
    else:
        main()