resizer_v5.py

import os
import time
from PIL import Image
from retinaface import RetinaFace
from concurrent.futures import ProcessPoolExecutor

Image.MAX_IMAGE_PIXELS = 200000000  # Allow images up to 200 million pixels

DATA = [
    ((768, 768), r"/kaggle/input/samikk"),
    ((1024, 1024), r"/kaggle/input/samikk"),
]



NUM_THREADS = 1  # Modify this for the desired number of threads
OUTPUT_ROOT = r"/kaggle/working/outputs"  # Root folder to save resized images
Save_PNG = False  # Options are True and False
LOG_FILE = "face_detection_failures.txt"  # File to log images where faces can't be detected

def resize_image(args):
    start_time = time.time()  # Start timing

    try:
        input_path, resolution, output_root, save_as_png = args
    except ValueError as e:
        print(f"Error unpacking arguments: {e}")
        return None

    try:
        # Load the image
        image = Image.open(input_path)
    except Exception as e:
        print(f"Error loading image: {e}")
        return None

    face_detected = True
    face_locations = None

    try:
        # Get face locations using RetinaFace
        faces = RetinaFace.detect_faces(input_path)

        if not faces or 'face_1' not in faces:
            face_detected = False
            with open(LOG_FILE, 'a') as log:
                log.write(f"Face not detected: {input_path}\n")
        else:
            face = faces['face_1']
            x, y, x2, y2 = face['facial_area']
            face_locations = [(y, x2, y2, x)]  # Adjust format to match face_recognition
    except Exception as e:
        print(f"Error detecting faces: {e}")
        face_detected = False

    # Proceed with cropping and resizing
    try:
        desired_aspect_ratio = resolution[0] / resolution[1]
        image_aspect_ratio = image.width / image.height

        # Calculate crop dimensions
        if image_aspect_ratio > desired_aspect_ratio:
            new_width = int(image.height * desired_aspect_ratio)
            new_height = image.height
        else:
            new_width = image.width
            new_height = int(image.width / desired_aspect_ratio)

        # Default centering
        left = (image.width - new_width) / 2
        top = (image.height - new_height) / 2
        right = (image.width + new_width) / 2
        bottom = (image.height + new_height) / 2

        # Adjust for face center if a face is detected
        if face_detected and face_locations:
            face_top, face_right, face_bottom, face_left = face_locations[0]
            face_center_x = (face_left + face_right) // 2
            face_center_y = (face_top + face_bottom) // 2

            left = min(max(0, face_center_x - new_width // 2), image.width - new_width)
            top = min(max(0, face_center_y - new_height // 2), image.height - new_height)
            right = left + new_width
            bottom = top + new_height

        image = image.crop((left, top, right, bottom))

        # Resize image with best resampling filter (LANCZOS)
        image = image.resize(resolution, Image.LANCZOS)

        output_folder = os.path.join(output_root, f"{resolution[0]}x{resolution[1]}")
        if not os.path.exists(output_folder):
            os.makedirs(output_folder)

        output_path = os.path.join(output_folder, os.path.basename(input_path))
        if save_as_png:
            output_path = os.path.splitext(output_path)[0] + '.png'
            image.save(output_path, format='PNG')
        else:
            image.save(output_path, quality=100)

    except Exception as e:
        print(f"Error processing or saving image: {e}")
        return None

    end_time = time.time()  # End timing
    return end_time - start_time  # Return processing time

def process_folder(input_folder, resolution, save_as_png):
    image_paths = [os.path.join(input_folder, fname) for fname in os.listdir(input_folder) if fname.lower().endswith(('png', 'jpg', 'jpeg'))]

    total_images = len(image_paths)
    processed_count = 0
    total_time = 0  # Total time for processing images

    with ProcessPoolExecutor(max_workers=NUM_THREADS) as executor:
        for processing_time in executor.map(resize_image, [(path, resolution, OUTPUT_ROOT, save_as_png) for path in image_paths]):
            processed_count += 1
            if processing_time is not None:
                total_time += processing_time
            average_time = total_time / processed_count * 1000  # Convert to milliseconds
            print(f"Processed {processed_count}/{total_images} images for resolution {resolution[0]}x{resolution[1]}... Average processing time per image: {average_time:.2f} ms")

if __name__ == "__main__":
    for resolution, folder in DATA:
        process_folder(folder, resolution, Save_PNG)
    print("Processing complete!")