utils/json_to_coco.py

import json
import os

def json_to_coco(config):
    base_dir = config["root"]
    image_dir = config["root_images"]
    annotations_dir = config["root_annotations"]

    def create_coco(json_dir):
        # COCO base structure
        coco = {
            "images": [],
            "annotations": [],
            "categories": []
        }
        uniq_filename_mapping = {}
        # Category mapping
        category_names = ["body", "face", "frame", "text"]
        category_map = {name: idx + 1 for idx, name in enumerate(category_names)}
        coco["categories"] = [{"id": cid, "name": name, "supercategory": "object"} for name, cid in category_map.items()]
        import copy
        coco_train = copy.deepcopy(coco)
        coco_test = copy.deepcopy(coco)
        coco_val = copy.deepcopy(coco)
        uniq = 0
        for book_json in os.listdir(json_dir):
            book = str(book_json).split(".")[0]
            data = []
            with open(base_dir +"json/"+book+".json", 'r',encoding='utf-8') as file:
                data = json.load(file)

            pages = data["page"]

            train_output = []
            val_output = []
            test_output = []

            import math
            import shutil

            page_count = len(pages) - 2
            div = page_count / 10
            div = int(math.ceil(div))
            # train_count = page_count - (2 * div)
            # val_count = div
            # test_count = div
            # print(train_count,val_count,test_count)
            train_arr = []
            val_arr = []
            test_arr = []
            count = 0
            for i in range(page_count,0,-1):
                if count < div :
                    test_arr.append(i)
                elif count < 2*div :
                    val_arr.append(i)
                else:
                    train_arr.append(i)
                count += 1

            # Annotation ID counter
            ann_id = 1

            # Helper to create COCO bbox annotation
            def create_annotation(ann, category_name, image_id, ann_id,uniq):
                x_min = int(ann["@xmin"])
                y_min = int(ann["@ymin"])
                x_max = int(ann["@xmax"])
                y_max = int(ann["@ymax"])
                width = x_max - x_min
                height = y_max - y_min

                return {
                    "id": ann_id,
                    "image_id": uniq,
                    "category_id": category_map[category_name],
                    "bbox": [x_min, y_min, width, height],
                    "area": width * height,
                    "iscrowd": 0
                }

            def append_data(coco, ann_id,uniq):
                # Add image entry
                coco["images"].append({
                    "id": uniq,
                    "width": page["@width"],
                    "height": page["@height"],
                    "file_name": f"{book}_{str(img_id).zfill(3)}.jpg"
                })


                # Loop through each category and add annotations
                for category in category_names:
                    for ann in page.get(category, []):
                        coco["annotations"].append(
                            create_annotation(ann, category, img_id, ann_id,uniq)
                        )
                        ann_id += 1

            
            for page in pages:
                uniq += 1
                img_id = int(page["@index"])
                if img_id == 0 or img_id == 1 :
                    continue

                curr_img_path = os.path.join(config["manga109_root_dir"],"images",book,f"{str(img_id).zfill(3)}.jpg")
                uniq_filename_mapping[uniq] = book+"_"+(curr_img_path.split("\\")[-1])
                if img_id in train_arr:
                    append_data(coco_train, ann_id,uniq)
                    train_output.append(coco)        
                    shutil.copyfile(curr_img_path,os.path.join(image_dir,"train/",book+"_"+(curr_img_path.split("\\")[-1])))
                if img_id in val_arr:
                    append_data(coco_val, ann_id,uniq)
                    val_output.append(coco)
                    shutil.copyfile(curr_img_path,os.path.join(image_dir,"val/",book+"_"+(curr_img_path.split("\\")[-1])))
                if img_id in test_arr:
                    append_data(coco_test, ann_id,uniq)
                    test_output.append(coco)
                    shutil.copyfile(curr_img_path,os.path.join(image_dir,"test/",book+"_"+(curr_img_path.split("\\")[-1])))

        with open(base_dir+"temp_uniq_file_name_mapping.json","w",encoding="utf-8") as f:
            json.dump(uniq_filename_mapping, f, ensure_ascii=False, indent=2)
        with open(annotations_dir+"instances_train.json", "w", encoding="utf-8") as f:
            json.dump(coco_train, f, ensure_ascii=False, indent=2)
        with open(annotations_dir+"instances_val.json", "w", encoding="utf-8") as f:
            json.dump(coco_val, f, ensure_ascii=False, indent=2)
        with open(annotations_dir+"instances_test.json", "w", encoding="utf-8") as f:
            json.dump(coco_test, f, ensure_ascii=False, indent=2)

    create_coco(base_dir+"json/")