Winows下pycococreator工具的使用---将自己的数据集转换为COCO类型

1. 安装winows版本的pycocotools工具

COCO 地址： https://github.com/cocodataset/cocoapi

大佬改写支持 Windows 的 COCO 地址：https://github.com/philferriere/cocoapi

 pip 安装:  pip install git+https://github.com/philferriere/cocoapi.git#subdirectory=PythonAPI

https://github.com/philferriere/cocoapi下载源码，并进行解压。以管理员身份打开 CMD 终端，并切换到 cocoapi\PythonAPI目录。运行以下指令

# install pycocotools locally

python setup.py build_ext --inplace

 # install pycocotools to the Python site-packages

python setup.py build_ext install

注意：Microsoft Visual C++ 14.0 is required

No module named ‘pycocotools._mask’： 工程下的pycocotools文件重名

 

2. 配置pycococreator工具

Github：https://github.com/waspinator/pycococreator/

安装命令: pip install git+git://github.com/waspinator/pycococreator.git@0.2.0

 

3.下载示例数据-形状数据集,放在合适路径

地址: https://patrickwasp.com/wp-content/uploads/2018/04/shapes_train_dataset.zip

4.shapes_to_coco.py代码

import datetime
import json
import os
import re
import fnmatch
from PIL import Image
import numpy as np
from pycococreatortools import pycococreatortools

ROOT_DIR = 'train'
IMAGE_DIR = os.path.join(ROOT_DIR, "shapes_train2018")
ANNOTATION_DIR = os.path.join(ROOT_DIR, "annotations")

INFO = {
    "description": "Example Dataset",
    "url": "https://github.com/waspinator/pycococreator",
    "version": "0.1.0",
    "year": 2018,
    "contributor": "waspinator",
    "date_created": datetime.datetime.utcnow().isoformat(' ')
}

LICENSES = [
    {
        "id": 1,
        "name": "Attribution-NonCommercial-ShareAlike License",
        "url": "http://creativecommons.org/licenses/by-nc-sa/2.0/"
    }
]

CATEGORIES = [
    {
        'id': 1,
        'name': 'square',
        'supercategory': 'shape',
    },
    {
        'id': 2,
        'name': 'circle',
        'supercategory': 'shape',
    },
    {
        'id': 3,
        'name': 'triangle',
        'supercategory': 'shape',
    },
]

def filter_for_jpeg(root, files):
    file_types = ['*.jpeg', '*.jpg']
    file_types = r'|'.join([fnmatch.translate(x) for x in file_types])
    files = [os.path.join(root, f) for f in files]
    files = [f for f in files if re.match(file_types, f)]
   
    return files

def filter_for_annotations(root, files, image_filename):
    file_types = ['*.png']
    file_types = r'|'.join([fnmatch.translate(x) for x in file_types])
    basename_no_extension = os.path.splitext(os.path.basename(image_filename))[0]
    file_name_prefix = basename_no_extension + '.*'
    files = [os.path.join(root, f) for f in files]
    files = [f for f in files if re.match(file_types, f)]
    files = [f for f in files if re.match(file_name_prefix, os.path.splitext(os.path.basename(f))[0])]

    return files

def main():

    coco_output = {
        "info": INFO,
        "licenses": LICENSES,
        "categories": CATEGORIES,
        "images": [],
        "annotations": []
    }

    image_id = 1
    segmentation_id = 1
   
    # filter for jpeg images
    for root, _, files in os.walk(IMAGE_DIR):
        image_files = filter_for_jpeg(root, files)

        # go through each image
        for image_filename in image_files:
            image = Image.open(image_filename)
            image_info = pycococreatortools.create_image_info(
                image_id, os.path.basename(image_filename), image.size)
            coco_output["images"].append(image_info)

            # filter for associated png annotations
            for root, _, files in os.walk(ANNOTATION_DIR):
                annotation_files = filter_for_annotations(root, files, image_filename)

                # go through each associated annotation
                for annotation_filename in annotation_files:
                   
                    print(annotation_filename)
                    class_id = [x['id'] for x in CATEGORIES if x['name'] in annotation_filename][0]

                    category_info = {'id': class_id, 'is_crowd': 'crowd' in image_filename}
                    binary_mask = np.asarray(Image.open(annotation_filename)
                        .convert('1')).astype(np.uint8)
                   
                    annotation_info = pycococreatortools.create_annotation_info(
                        segmentation_id, image_id, category_info, binary_mask,
                        image.size, tolerance=2)

                    if annotation_info is not None:
                        coco_output["annotations"].append(annotation_info)

                    segmentation_id = segmentation_id + 1

            image_id = image_id + 1

    with open('{}/instances_shape_train2018.json'.format(ROOT_DIR), 'w') as output_json_file:
        json.dump(coco_output, output_json_file)


if __name__ == "__main__":
    main()