coco_to_yolo.py

import os
import json
from PIL import Image

def yolo_to_coco(yolo_bbox, img_width, img_height):
    """
    Convert YOLO format bounding box to COCO format.
    YOLO format: (x_center, y_center, width, height) normalized [0, 1]
    COCO format: [x_min, y_min, width, height] in pixels
    """
    x_center, y_center, width, height = yolo_bbox
    # Convert normalized values to pixel values
    x_center *= img_width
    y_center *= img_height
    width *= img_width
    height *= img_height
    # Calculate x_min and y_min
    x_min = x_center - width / 2
    y_min = y_center - height / 2
    return [x_min, y_min, width, height]

def convert_to_coco_format(image_dir, label_dir, output_json, categories):
    """
    Convert a dataset from YOLO format to COCO format.
    
    Parameters:
    - image_dir: Directory containing images
    - label_dir: Directory containing YOLO format label files
    - output_json: Path to save the output COCO format JSON file
    - categories: List of category dictionaries for COCO format
    """
    images = []
    annotations = []
    annotation_id = 1

    for image_file in os.listdir(image_dir):
        if not image_file.endswith(('.jpg', '.png')):
            continue
        
        image_id = len(images) + 1
        img_path = os.path.join(image_dir, image_file)
        img = Image.open(img_path)
        width, height = img.size
        
        images.append({
            "id": image_id,
            "width": width,
            "height": height,
            "file_name": image_file
        })

        label_file = os.path.join(label_dir, os.path.splitext(image_file)[0] + '.txt')
        if not os.path.exists(label_file):
            continue
        
        with open(label_file) as f:
            for line in f:
                # Read class_id and bounding box coordinates from YOLO label file
                class_id, x_center, y_center, bbox_width, bbox_height = map(float, line.strip().split())
                # Convert YOLO bounding box to COCO bounding box
                bbox = yolo_to_coco([x_center, y_center, bbox_width, bbox_height], width, height)
                
                annotations.append({
                    "id": annotation_id,
                    "image_id": image_id,
                    "category_id": int(class_id) + 1,
                    "bbox": bbox,
                    "area": bbox[2] * bbox[3],  # area = width * height
                    "iscrowd": 0,
                    "segmentation": []  # No segmentation info available
                })
                annotation_id += 1

    coco_format = {
        "images": images,
        "annotations": annotations,
        "categories": categories
    }

    with open(output_json, 'w') as f:
        json.dump(coco_format, f, indent=4)

# List of class names for the dataset
class_names = ['airan-katyk', 'almond', 'apple', 'artichoke', 'arugula', 'asparagus', 'avocado', 'bacon', 'banana', 'beans', 'beet', 'bell pepper', 'black olives', 'blackberry', 'blueberry', 'boiled chicken', 'bread', 'broccoli', 'buckwheat', 'cabbage', 'cakes', 'carrot', 'cashew', 'casserole with meat and vegetables', 'cauliflower', 'celery', 'cereal based cooked food', 'cheese', 'chickpeas', 'chips', 'cooked eggplant', 'cooked food based on meat', 'cooked food meat with vegetables', 'cooked zucchini', 'cookies', 'corn', 'crepe', 'cucumber', 'cutlet', 'desserts', 'egg product', 'eggplant', 'fish', 'fried chicken', 'fried eggs', 'fried fish', 'fried meat', 'fruits', 'granola', 'grapes', 'green beans', 'herbs', 'hummus', 'ice-cream', 'irimshik', 'juice', 'kiwi', 'lavash', 'legumes', 'lemon', 'mandarin', 'mango', 'mashed potato', 'meat product', 'melon', 'mixed berries', 'mixed nuts', 'mushrooms', 'onion', 'orange', 'pasta', 'pastry', 'peanut', 'pear', 'peas', 'pecan', 'pickled cabbage', 'pickled squash', 'pie', 'pineapple', 'pizza', 'plov', 'porridge', 'potatoes', 'pumpkin', 'radish', 'raspberry', 'rice', 'salad fresh', 'salad leaves', 'salad with fried meat veggie', 'salad with sauce', 'sandwich', 'sausages', 'seafood', 'smetana', 'snacks', 'snacks bread', 'souces', 'soup-plain', 'soy product', 'spinach', 'strawberry', 'suzbe', 'sweet potatoes', 'tomato', 'tomato souce', 'tushpara-wo-soup', 'vegetable based cooked food', 'waffles', 'walnut', 'watermelon', 'zucchini']
assert len(class_names) == 113
categories = [{"id": i + 1, "name": name, "supercategory": "none"} for i, name in enumerate(class_names)]

# Directory paths for the dataset
dataset_dir = '../datasets/Nutrition5k'
output_dir = '../datasets/Nutrition5k/annotations'
os.makedirs(output_dir, exist_ok=True)

# Convert labels for train, validation, and test splits
for split in ['train', 'val', 'test']:
    image_dir = os.path.join(dataset_dir, split, 'images')
    label_dir = os.path.join(dataset_dir, split, 'labels')
    output_json = os.path.join(output_dir, f'instances_{split}.json')
    convert_to_coco_format(image_dir, label_dir, output_json, categories)