tools/conversion/obb2coco.py

import json
import numpy as np
import os
import cv2
from datetime import datetime

def convert_to_coco(yolo_obb_dir, image_dir, save_json, categories, split="Train", author="None", version="1.0"):
    """
    yolo_obb_dir: Path to .txt files (class_id x1 y1 x2 y2 x3 y3 x4 y4)
    image_dir: Path to corresponding images
    save_json: Path to output coco file
    categories: List of strings ['bubble', ...]
    """
    
    coco_output = {
        "info": {
            "description": "TinyBubble Dataset",
            "version": version,
            "year": datetime.now().year,
            "split": split
        },
        "images": [],
        "annotations": [],
        "categories": [
            {"id": i, "name": cat, "supercategory": "none"} 
            for i, cat in enumerate(categories)
        ]
    }

    ann_id = 0
    img_id = 0

    if not os.path.exists(yolo_obb_dir):
        print(f"Directory {yolo_obb_dir} not found.")
        return

    for filename in os.listdir(yolo_obb_dir):
        if not filename.endswith(".txt"): continue
        
        img_base = os.path.splitext(filename)[0]
        image_name = None
        width, height = 0, 0
        
        for ext in ['.png', '.jpg', '.jpeg']:
            temp_path = os.path.join(image_dir, img_base + ext)
            if os.path.exists(temp_path):
                img = cv2.imread(temp_path)
                if img is not None:
                    height, width, _ = img.shape
                    image_name = img_base + ext
                    break
        
        if image_name is None:
            print(f"Warning: Image for {filename} not found in {image_dir}. Skipping.")
            continue

        coco_output["images"].append({
            "id": img_id,
            "file_name": image_name,
            "width": width,  
            "height": height 
        })

        with open(os.path.join(yolo_obb_dir, filename), "r") as f:
            lines = list(dict.fromkeys([line.strip() for line in f.readlines() if line.strip()]))
            
            for line in lines:
                parts = list(map(float, line.split()))
                class_id = int(parts[0])
                obb_coords = parts[1:]
                
                abs_coords = []
                for i in range(0, len(obb_coords), 2):
                    abs_coords.append(float(obb_coords[i] * width))
                    abs_coords.append(float(obb_coords[i+1] * height))

                points = np.array(abs_coords).reshape(4, 2)
                x_min, y_min = np.min(points, axis=0)
                x_max, y_max = np.max(points, axis=0)
                
                side_a = np.linalg.norm(points[0] - points[1])
                side_b = np.linalg.norm(points[1] - points[2])

                ann = {
                    "id": ann_id,
                    "image_id": img_id,
                    "category_id": class_id,
                    "segmentation": [abs_coords],
                    "area": float(side_a * side_b),
                    "bbox": [float(x_min), float(y_min), float(x_max - x_min), float(y_max - y_min)],
                    "iscrowd": 0
                }
                coco_output["annotations"].append(ann)
                ann_id += 1
        
        img_id += 1

    with open(save_json, "w") as f:
        json.dump(coco_output, f, indent=4)
    
    print(f"Successfully cleaned and saved {ann_id-1} annotations to {save_json}")

convert_to_coco(
    '../../../tinybubble/yolo_obb/train/labels', 
    '../../../tinybubble/yolo_obb/train/images', 
    '../../../tinybubble/coco/annotations/1.0_train_coco_obb.json', 
    ['bubble'],
    split="Train",
    version="1.0"
)

convert_to_coco(
    '../../../tinybubble/yolo_obb/val/labels', 
    '../../../tinybubble/yolo_obb/val/images', 
    '../../../tinybubble/coco/annotations/1.0_val_coco_obb.json', 
    ['bubble'],
    split="Val",
    version="1.0"
)