label_converter.py

import argparse
import json
import os
import os.path as osp
import time

from PIL import Image
from tqdm import tqdm
from datetime import date

import numpy as np
import xml.dom.minidom as minidom
import xml.etree.ElementTree as ET

import sys
sys.path.append('.')
from anylabeling.app_info import __version__

VERSION = __version__


class BaseLabelConverter:
    def __init__(self, classes_file=None):

        if classes_file:
            with open(classes_file, 'r', encoding='utf-8') as f:
                self.classes = f.read().splitlines()
        else:
            self.classes = []
        print(f"import classes is: {self.classes}")

    def reset(self):
        self.custom_data = dict(
            version=VERSION,
            flags={},
            shapes=[],
            imagePath="",
            imageData=None,
            imageHeight=-1,
            imageWidth=-1
        )

    def get_image_size(self, image_file):
        with Image.open(image_file) as img:
            width, height = img.size
            return width, height

    def get_minimal_enclosing_rectangle(self, poly):
        assert len(poly) == 8, \
            "Input rectangle must contain exactly 8 values."
        
        x_coords = [poly[i] for i in range(0, 8, 2)]
        y_coords = [poly[i] for i in range(1, 8, 2)]

        min_x = min(x_coords)
        max_x = max(x_coords)
        min_y = min(y_coords)
        max_y = max(y_coords)

        x = min_x
        y = min_y
        width = max_x - min_x
        height = max_y - min_y

        return [x, y, width, height]

    def get_poly_area(self, poly):
        # Ensure that poly contains exactly 8 values
        assert len(poly) == 8, \
            "Input polygon must contain exactly 8 values."

        # Extract x and y coordinates
        x_coords = [poly[i] for i in range(0, 8, 2)]
        y_coords = [poly[i] for i in range(1, 8, 2)]

        # Calculate the area using the Shoelace formula
        area = 0.5 * abs(
            sum(
                x_coords[i] * y_coords[i + 1] - x_coords[i + 1] * y_coords[i]
                for i in range(3)
            ) + x_coords[3] * y_coords[0] - x_coords[0] * y_coords[3]
        )

        return area

    def get_coco_data(self):
        coco_data = {
            "info": {
                "year": 2023,
                "version": VERSION,
                "description": "COCO Label Conversion",
                "contributor": "CVHub",
                "url": "https://github.com/CVHub520/X-AnyLabeling",
                "date_created": str(date.today())
            },
            "licenses": [
                {
                    "id": 1,
                    "url": "https://www.gnu.org/licenses/gpl-3.0.html",
                    "name": "GNU GENERAL PUBLIC LICENSE Version 3"
                }
            ],
            "categories": [],
            "images": [],
            "annotations": []
        }
        return coco_data

    def ensure_output_path(self, output_path, ext=None):
        if osp.isfile(output_path):
            # Check if the file has the expected extension
            if not output_path.lower().endswith(ext.lower()):
                raise ValueError(
                    f"The specified file '{output_path}' \
                        does not have the expected '{ext}' extension."
                )
        else:
            # Check if the folder exists, and create it if it doesn't
            if not osp.exists(output_path):
                os.makedirs(output_path, exist_ok=True)

class RectLabelConverter(BaseLabelConverter):

    def custom_to_voc2017(self, input_file, output_dir):
        with open(input_file, 'r', encoding='utf-8') as f:
            data = json.load(f)

        image_path = data['imagePath']
        image_width = data['imageWidth']
        image_height = data['imageHeight']

        root = ET.Element('annotation')
        ET.SubElement(root, 'folder').text = osp.dirname(output_dir)
        ET.SubElement(root, 'filename').text = osp.basename(image_path)
        size = ET.SubElement(root, 'size')
        ET.SubElement(size, 'width').text = str(image_width)
        ET.SubElement(size, 'height').text = str(image_height)
        ET.SubElement(size, 'depth').text = '3'

        for shape in data['shapes']:
            label = shape['label']
            points = shape['points']

            xmin = str(points[0][0])
            ymin = str(points[0][1])
            xmax = str(points[1][0])
            ymax = str(points[1][1])

            object_elem = ET.SubElement(root, 'object')
            ET.SubElement(object_elem, 'name').text = label
            ET.SubElement(object_elem, 'pose').text = 'Unspecified'
            ET.SubElement(object_elem, 'truncated').text = '0'
            ET.SubElement(object_elem, 'difficult').text = '0'
            bndbox = ET.SubElement(object_elem, 'bndbox')
            ET.SubElement(bndbox, 'xmin').text = xmin
            ET.SubElement(bndbox, 'ymin').text = ymin
            ET.SubElement(bndbox, 'xmax').text = xmax
            ET.SubElement(bndbox, 'ymax').text = ymax

        xml_string = ET.tostring(root, encoding='utf-8')
        dom = minidom.parseString(xml_string)
        formatted_xml = dom.toprettyxml(indent='  ')

        with open(output_dir, 'w') as f:
            f.write(formatted_xml)

    def voc2017_to_custom(self, input_file, output_file):
        self.reset()

        tree = ET.parse(input_file)
        root = tree.getroot()

        image_path = root.find('filename').text
        image_width = int(root.find('size/width').text)
        image_height = int(root.find('size/height').text)

        self.custom_data['imagePath'] = image_path
        self.custom_data['imageHeight'] = image_height
        self.custom_data['imageWidth'] = image_width

        for obj in root.findall('object'):
            label = obj.find('name').text
            xmin = float(obj.find('bndbox/xmin').text)
            ymin = float(obj.find('bndbox/ymin').text)
            xmax = float(obj.find('bndbox/xmax').text)
            ymax = float(obj.find('bndbox/ymax').text)

            shape = {
                "label": label,
                "text": "",
                "points": [[xmin, ymin], [xmax, ymax]],
                "group_id": None,
                "shape_type": "rectangle",
                "flags": {}
            }

            self.custom_data['shapes'].append(shape)

        with open(output_file, 'w', encoding='utf-8') as f:
            json.dump(self.custom_data, f, indent=2, ensure_ascii=False)

    def custom_to_yolov5(self, input_file, output_file):
        with open(input_file, 'r', encoding='utf-8') as f:
            data = json.load(f)

        image_width = data['imageWidth']
        image_height = data['imageHeight']

        with open(output_file, 'w', encoding='utf-8') as f:
            for shape in data['shapes']:
                label = shape['label']
                points = shape['points']

                class_index = self.classes.index(label)

                x_center = (points[0][0] + points[1][0]) / (2 * image_width)
                y_center = (points[0][1] + points[1][1]) / (2 * image_height)
                width = abs(points[1][0] - points[0][0]) / image_width
                height = abs(points[1][1] - points[0][1]) / image_height

                f.write(f"{class_index} {x_center} {y_center} {width} {height}\n")

    def yolov5_to_custom(self, input_file, output_file, image_file):
        self.reset()

        with open(input_file, 'r', encoding='utf-8') as f:
            lines = f.readlines()

        image_width, image_height = self.get_image_size(image_file)

        for line in lines:
            line = line.strip().split(' ')
            class_index = int(line[0])
            x_center = float(line[1])
            y_center = float(line[2])
            width = float(line[3])
            height = float(line[4])

            x_min = int((x_center - width / 2) * image_width)
            y_min = int((y_center - height / 2) * image_height)
            x_max = int((x_center + width / 2) * image_width)
            y_max = int((y_center + height / 2) * image_height)

            label = self.classes[class_index]

            shape = {
                "label": label,
                "text": None,
                "points": [[x_min, y_min], [x_max, y_max]],
                "group_id": None,
                "shape_type": "rectangle",
                "flags": {}
            }

            self.custom_data['shapes'].append(shape)

        self.custom_data['imagePath'] = osp.basename(image_file)
        self.custom_data['imageHeight'] = image_height
        self.custom_data['imageWidth'] = image_width

        with open(output_file, 'w', encoding='utf-8') as f:
            json.dump(self.custom_data, f, indent=2, ensure_ascii=False)

    def custom_to_coco(self, input_path, output_path):
        coco_data = self.get_coco_data()

        for i, class_name in enumerate(self.classes):
            coco_data['categories'].append({
                "id": i+1,
                "name": class_name,
                "supercategory": ""
            })

        image_id = 0
        annotation_id = 0

        file_list = os.listdir(input_path)
        for file_name in tqdm(file_list, desc='Converting files', unit='file', colour='green'):
            if not file_name.endswith('.json'): continue
            image_id += 1

            input_file = osp.join(input_path, file_name)
            with open(input_file, 'r', encoding='utf-8') as f:
                data = json.load(f)

            image_path = data['imagePath']
            image_name = osp.splitext(osp.basename(image_path))[0]

            coco_data['images'].append({
                "id": image_id,
                "file_name": image_name,
                "width": data['imageWidth'],
                "height": data['imageHeight'],
                "license": 0,
                "flickr_url": "",
                "coco_url": "",
                "date_captured": ""
            })

            for shape in data['shapes']:
                annotation_id += 1
                label = shape['label']
                points = shape['points']
                class_id = self.classes.index(label)
                x_min = min(points[0][0], points[1][0])
                y_min = min(points[0][1], points[1][1])
                x_max = max(points[0][0], points[1][0])
                y_max = max(points[0][1], points[1][1])
                width = x_max - x_min
                height = y_max - y_min

                annotation = {
                    "id": annotation_id,
                    "image_id": image_id,
                    "category_id": class_id+1,
                    "bbox": [x_min, y_min, width, height],
                    "area": width * height,
                    "iscrowd": 0
                }

                coco_data['annotations'].append(annotation)

        output_file = osp.join(output_path, "x_anylabeling_coco.json")
        with open(output_file, 'w', encoding='utf-8') as f:
            json.dump(coco_data, f, indent=4, ensure_ascii=False)

    def coco_to_custom(self, input_file, output_path, image_path):

        img_dic = {}
        for file in os.listdir(image_path):
            img_dic[file] = file

        with open(input_file, 'r', encoding='utf-8') as f:
            data = json.load(f)
            
        if not self.classes:
            for cat in data["categories"]:
                self.classes.append(cat["name"])

        total_info, label_info = {}, {}

        # map category_id to name
        for dic_info in data["categories"]:
            label_info[dic_info["id"]] = dic_info["name"]

        # map image_id to info
        for dic_info in data["images"]:
            total_info[dic_info["id"]] = {
                "imageWidth": dic_info["width"],
                "imageHeight": dic_info["height"],
                "imagePath": img_dic[dic_info["file_name"]],
                "shapes": []
            }
        
        for dic_info in data["annotations"]:

            bbox = dic_info["bbox"]
            x_min = bbox[0]
            y_min = bbox[1]
            width = bbox[2]
            height = bbox[3]
            x_max = x_min + width
            y_max = y_min + height

            shape_info = {
                "label": self.classes[dic_info["category_id"]-1],
                "text": None,
                "points": [[x_min, y_min], [x_max, y_max]],
                "group_id": None,
                "shape_type": "rectangle",
                "flags": {}
            }

            total_info[dic_info["image_id"]]["shapes"].append(shape_info)
    
        for dic_info in tqdm(total_info.values(), desc='Converting files', unit='file', colour='green'):
            self.reset()
            self.custom_data["shapes"] = dic_info["shapes"]
            self.custom_data["imagePath"] = dic_info["imagePath"]
            self.custom_data["imageHeight"] = dic_info["imageHeight"]
            self.custom_data["imageWidth"] = dic_info["imageWidth"]

            output_file = osp.join(output_path, osp.splitext(dic_info["imagePath"])[0]+".json")
            with open(output_file, 'w', encoding='utf-8') as f:
                json.dump(self.custom_data, f, indent=2, ensure_ascii=False)

class PolyLabelConvert(BaseLabelConverter):

    def custom_to_yolov5(self, input_file, output_file):
        with open(input_file, 'r', encoding='utf-8') as f:
            data = json.load(f)

        image_width = data['imageWidth']
        image_height = data['imageHeight']
        image_size = np.array([[image_width, image_height]])

        with open(output_file, 'w', encoding='utf-8') as f:
            for shape in data['shapes']:
                label = shape['label']
                points = np.array(shape['points'])
                class_index = self.classes.index(label)
                norm_points = points / image_size
                f.write(f"{class_index} " + " ".join([" ".join([str(cell[0]), str(cell[1])]) for cell in norm_points.tolist()]) + "\n")

    def yolov5_to_custom(self, input_file, output_file, image_file):
        self.reset()

        with open(input_file, 'r', encoding='utf-8') as f:
            lines = f.readlines()

        image_width, image_height = self.get_image_size(image_file)
        image_size = np.array([image_width, image_height], np.float64)

        for line in lines:
            line = line.strip().split(' ')
            class_index = int(line[0])
            label = self.classes[class_index]
            masks = line[1:]
            shape = {
                "label": label,
                "text": None,
                "points": [],
                "group_id": None,
                "shape_type": "rectangle",
                "flags": {}
            }
            for x, y in zip(masks[0::2], masks[1::2]):
                point = [np.float64(x), np.float64(y)]
                point = np.array(point, np.float64) * image_size
                shape['points'].append(point.tolist())
            self.custom_data['shapes'].append(shape)

        self.custom_data['imagePath'] = osp.basename(image_file)
        self.custom_data['imageHeight'] = image_height
        self.custom_data['imageWidth'] = image_width

        with open(output_file, 'w', encoding='utf-8') as f:
            json.dump(self.custom_data, f, indent=2, ensure_ascii=False)

class RotateLabelConverter(BaseLabelConverter):

    def custom_to_dota(self, data, output_file):
        with open(output_file, 'w', encoding='utf-8') as f:
            for shape in data['shapes']:
                label = shape['label']
                points = shape['points']
                x0 = points[0][0]
                y0 = points[0][1]
                x1 = points[1][0]
                y1 = points[1][1]
                x2 = points[2][0]
                y2 = points[2][1]
                x3 = points[3][0]
                y3 = points[3][1]
                f.write(f"{x0} {y0} {x1} {y1} {x2} {y2} {x3} {y3} {label} 0\n")

    def dota_to_custom(self, input_file, output_file, image_file):
        self.reset()

        with open(input_file, 'r', encoding='utf-8') as f:
            lines = f.readlines()

        image_width, image_height = self.get_image_size(image_file)

        for line in lines:
            line = line.strip().split(' ')
            x0, y0, x1, y1, x2, y2, x3, y3 = [float(i) for i in line[:8]]
            shape = {
                "label": line[8],
                "text": None,
                "points": [
                    [x0, y0],
                    [x1, y1],
                    [x2, y2],
                    [x3, y3]
                ],
                "group_id": None,
                "direction": 0,
                "shape_type": "rotation",
                "flags": {}
            }
            self.custom_data['shapes'].append(shape)

        self.custom_data['imagePath'] = osp.basename(image_file)
        self.custom_data['imageHeight'] = image_height
        self.custom_data['imageWidth'] = image_width

        with open(output_file, 'w', encoding='utf-8') as f:
            json.dump(self.custom_data, f, indent=2, ensure_ascii=False)

    def dota_to_dcoco(self, input_path, output_path, image_path):
        self.ensure_output_path(output_path, 'json')
        coco_data = self.get_coco_data()

        for i, class_name in enumerate(self.classes):
            coco_data['categories'].append({
                "id": i+1,
                "name": class_name,
                "supercategory": ""
            })

        image_id = 0
        annotation_id = 0

        file_list = os.listdir(image_path)
        for image_file in tqdm(
            file_list, desc='Converting files', unit='file', colour='green'
        ):
            label_file = osp.join(input_path, osp.splitext(image_file)[0]+'.txt')

            image_width, image_height = self.get_image_size(
                osp.join(image_path, image_file)
            )
            
            image_id += 1

            coco_data['images'].append({
                "id": image_id,
                "file_name": image_file,
                "width": image_width,
                "height": image_height,
                "license": 0,
                "flickr_url": "",
                "coco_url": "",
                "date_captured": ""
            })

            with open(label_file, 'r', encoding='utf-8') as f:
                lines = f.readlines()
            
            for line in lines:
                line = line.strip().split(' ')
                *poly, label, _ = line
                poly = list(map(float, poly))
                area = self.get_poly_area(poly)
                rect = self.get_minimal_enclosing_rectangle(poly)
                annotation_id += 1
                class_id = self.classes.index(label)
                annotation = {
                    "id": annotation_id,
                    "image_id": image_id,
                    "category_id": class_id+1,
                    "bbox": rect,
                    "segmentation": [poly],
                    "area": area,
                    "iscrowd": 0
                }

                coco_data['annotations'].append(annotation)

        if osp.isdir(output_path):
            output_path = osp.join(output_path, "x_anylabeling_coco.json")

        with open(output_path, 'w', encoding='utf-8') as f:
            json.dump(coco_data, f, indent=4, ensure_ascii=False) 

    def dcoco_to_dota(self, input_file, output_path):
        self.ensure_output_path(output_path)
        with open(input_file, 'r', encoding='utf-8') as f:
            data = json.load(f)
        
        label_info = {}
        # map category_id to label
        for dic_info in data["categories"]:
            label_info[dic_info["id"]] = dic_info["name"]

        name_info = {}
        # map image_id to file_naame
        for dic_info in data["images"]:
            name_info[dic_info["id"]] = dic_info["file_name"]

        total_info = {}
        for dic_info in data["annotations"]:
            poly = dic_info["segmentation"][0]
            image_id = dic_info["image_id"]
            category_id = dic_info["category_id"]
            label = label_info[category_id]
            if image_id not in total_info:
                total_info[image_id] = [[*poly, label]]
            else:
                total_info[image_id].append([*poly, label])
        
        for image_id, label_info in total_info.items():
            label_file = osp.basename(name_info[image_id]) + '.txt'
            output_file = osp.join(output_path, label_file)
            with open(output_file, 'w', encoding='utf-8') as f:
                for info in label_info:
                    x0, y0, x1, y1, x2, y2, x3, y3, label = info
                    f.write(f"{x0} {y0} {x1} {y1} {x2} {y2} {x3} {y3} {label} 0\n")  

    def dcoco_to_yolo(self, input_file, output_path):
        self.ensure_output_path(output_path)
        with open(input_file, 'r', encoding='utf-8') as f:
            data = json.load(f)
        
        label_info = {}
        # map category_id to label
        for dic_info in data["categories"]:
            label_info[dic_info["id"]] = dic_info["id"] - 1

        name_info = {}
        # map image_id to file_naame
        for dic_info in data["images"]:
            name_info[dic_info["id"]] = dic_info["file_name"]

        total_info = {}
        for dic_info in data["annotations"]:
            poly = dic_info["segmentation"][0]
            image_id = dic_info["image_id"]
            category_id = dic_info["category_id"]
            label = label_info[category_id]
            if image_id not in total_info:
                total_info[image_id] = [[*poly, label]]
            else:
                total_info[image_id].append([*poly, label])
        
        for image_id, label_info in total_info.items():
            label_file = osp.basename(name_info[image_id]) + '.txt'
            output_file = osp.join(output_path, label_file)
            with open(output_file, 'w', encoding='utf-8') as f:
                for info in label_info:
                    x0, y0, x1, y1, x2, y2, x3, y3, label = info
                    f.write(f"{x0} {y0} {x1} {y1} {x2} {y2} {x3} {y3} {label} 0\n")  

def main():
    parser = argparse.ArgumentParser(description='Label Converter')
    parser.add_argument('--task', default='rectangle', choices=['rectangle', 'polygon', 'rotation'], help='Choose the type of task to perform')
    parser.add_argument('--src_path', help='Path to input directory')
    parser.add_argument('--dst_path', help='Path to output directory')
    parser.add_argument('--img_path', help='Path to image directory')
    parser.add_argument('--classes', default=None, help='Path to classes.txt file, where each line represent a specific class')
    parser.add_argument('--mode', help='Choose the conversion mode what you need',
                        choices=['custom2voc', 'voc2custom', 'custom2yolo', 'yolo2custom', 'custom2coco', 'coco2custom', 
                                 'custom2dota', 'dota2custom', 'dota2dcoco', 'dcoco2dota'])
    args = parser.parse_args()

    print(f"Starting conversion to {args.mode} format of {args.task}...")
    start_time = time.time()

    if args.task == 'rectangle':
        converter = RectLabelConverter(args.classes)
    elif args.task == 'polygon':
        converter = PolyLabelConvert(args.classes)
        valid_modes = ['custom2yolo', 'yolo2custom']
        assert args.mode in valid_modes, f"Polygon tasks are only supported in {valid_modes} now!"
    elif args.task == "rotation":
        converter = RotateLabelConverter(args.classes)
        valid_modes = ['custom2dota', 'dota2custom', 'dota2dcoco', 'dcoco2dota']
        assert args.mode in valid_modes, f"Rotation tasks are only supported in {valid_modes} now!"

    if args.mode == "custom2voc":
        file_list = os.listdir(args.src_path)
        os.makedirs(args.dst_path, exist_ok=True)
        for file_name in tqdm(file_list, desc='Converting files', unit='file', colour='green'):
            if not file_name.endswith('.json'): continue
            src_file = osp.join(args.src_path, file_name)
            dst_file = osp.join(args.dst_path, osp.splitext(file_name)[0]+'.xml')
            converter.custom_to_voc2017(src_file, dst_file)
    elif args.mode == "voc2custom":
        file_list = os.listdir(args.src_path)
        for file_name in tqdm(file_list, desc='Converting files', unit='file', colour='green'):
            src_file = osp.join(args.src_path, file_name)
            dst_file = osp.join(args.img_path, osp.splitext(file_name)[0]+'.json')
            converter.voc2017_to_custom(src_file, dst_file)
    elif args.mode == "custom2yolo":
        file_list = os.listdir(args.src_path)
        os.makedirs(args.dst_path, exist_ok=True)
        for file_name in tqdm(file_list, desc='Converting files', unit='file', colour='green'):
            if not file_name.endswith('.json'): continue
            src_file = osp.join(args.src_path, file_name)
            dst_file = osp.join(args.dst_path, osp.splitext(file_name)[0]+'.txt')
            converter.custom_to_yolov5(src_file, dst_file)
    elif args.mode == "yolo2custom":
        img_dic = {}
        for file in os.listdir(args.img_path):
            prefix = file.rsplit('.', 1)[0]
            img_dic[prefix] = file
        file_list = os.listdir(args.src_path)
        for file_name in tqdm(file_list, desc='Converting files', unit='file', colour='green'):
            src_file = osp.join(args.src_path, file_name)
            dst_file = osp.join(args.img_path, osp.splitext(file_name)[0]+'.json')
            img_file = osp.join(args.img_path, img_dic[osp.splitext(file_name)[0]])
            converter.yolov5_to_custom(src_file, dst_file, img_file)
    elif args.mode == "custom2coco":
        os.makedirs(args.dst_path, exist_ok=True)
        converter.custom_to_coco(args.src_path, args.dst_path)
    elif args.mode == "coco2custom":
        os.makedirs(args.dst_path, exist_ok=True)
        converter.coco_to_custom(args.src_path, args.dst_path, args.img_path)
    elif args.mode == "custom2dota":
        file_list = os.listdir(args.src_path)
        os.makedirs(args.dst_path, exist_ok=True)
        for file_name in tqdm(file_list, desc='Converting files', unit='file', colour='green'):
            if not file_name.endswith('.json'): continue
            src_file = osp.join(args.src_path, file_name)
            dst_file = osp.join(args.dst_path, osp.splitext(file_name)[0]+'.txt')
            converter.custom_to_dota(src_file, dst_file)
    elif args.mode == "dota2custom":
        img_dic = {}
        for file in os.listdir(args.img_path):
            prefix = file.rsplit('.', 1)[0]
            img_dic[prefix] = file
        file_list = os.listdir(args.src_path)
        for file_name in tqdm(file_list, desc='Converting files', unit='file', colour='green'):
            src_file = osp.join(args.src_path, file_name)
            dst_file = osp.join(args.img_path, osp.splitext(file_name)[0]+'.json')
            img_file = osp.join(args.img_path, img_dic[osp.splitext(file_name)[0]])
            converter.dota_to_custom(src_file, dst_file, img_file)
    elif args.mode == "dota2dcoco":
        converter.dota_to_dcoco(args.src_path, args.dst_path, args.img_path)
    elif args.mode == "dcoco2dota":
        converter.dcoco_to_dota(args.src_path, args.dst_path)

    end_time = time.time()
    print(f"Conversion completed successfully: {args.dst_path}")
    print(f"Conversion time: {end_time - start_time:.2f} seconds")

if __name__ == '__main__':
    main()