[自用代码]labelme--人脸关键点标注--json转xml--xml转txt

1. labelme标注人脸：

(翻个白眼先)用“Create rectangle”和“Create Point”，类别分别为“face, le, re, no, lm, rm”(脸，左眼，右眼，鼻子，左嘴角，右嘴角)；

标注好后会生成json文件内容具体如下：

{
  "version": "5.3.1",
  "flags": {},
  "shapes": [
    {
      "label": "face",
      "points": [
        [
          1222.1052631578948,
          447.4436090225564
        ],
        [
          1677.7443609022555,
          1112.1052631578948
        ]
      ],
      "group_id": null,
      "description": "",
      "shape_type": "rectangle",
      "flags": {}
    },
    {
      "label": "le",
      "points": [
        [
          1383.7593984962405,
          757.9699248120301
        ]
      ],
      "group_id": null,
      "description": "",
      "shape_type": "point",
      "flags": {}
    },
    {
      "label": "re",
      "points": [
        [
          1558.1954887218044,
          758.7218045112782
        ]
      ],
      "group_id": null,
      "description": "",
      "shape_type": "point",
      "flags": {}
    },
    {
      "label": "no",
      "points": [
        [
          1477.7443609022555,
          864.7368421052631
        ]
      ],
      "group_id": null,
      "description": "",
      "shape_type": "point",
      "flags": {}
    },
    {
      "label": "lm",
      "points": [
        [
          1400.3007518796992,
          979.7744360902255
        ]
      ],
      "group_id": null,
      "description": "",
      "shape_type": "point",
      "flags": {}
    },
    {
      "label": "rm",
      "points": [
        [
          1540.9022556390976,
          979.7744360902255
        ]
      ],
      "group_id": null,
      "description": "",
      "shape_type": "point",
      "flags": {}
    }
  ],
  "imagePath": "WIN_20231224_10_14_05_Pro.jpg",

2 解析 json 文件

生成VOC格式的文件夹，参考一些代码，进行修改，实现对“point”和“bndbox”类别的读取，并生成xml文件，运行命令：python labelme2voc.py temp_face wider --label label.txt

#!/usr/bin/env python

from __future__ import print_function

import argparse
import glob
import os
import os.path as osp
import sys

import imgviz
import labelme

try:
    import lxml.builder
    import lxml.etree
except ImportError:
    print("Please install lxml:\n\n    pip install lxml\n")
    sys.exit(1)


def main():
    parser = argparse.ArgumentParser(
        formatter_class=argparse.ArgumentDefaultsHelpFormatter
    )
    parser.add_argument("input_dir", help="input annotated directory")
    parser.add_argument("output_dir", help="output dataset directory")
    parser.add_argument("--labels", help="labels file", required=True)
    parser.add_argument(
        "--noviz", help="no visualization", action="store_true"
    )
    args = parser.parse_args()

    if osp.exists(args.output_dir):
        print("Output directory already exists:", args.output_dir)
        sys.exit(1)
    os.makedirs(args.output_dir)
    os.makedirs(osp.join(args.output_dir, "JPEGImages"))
    os.makedirs(osp.join(args.output_dir, "Annotations"))
    if not args.noviz:
        os.makedirs(osp.join(args.output_dir, "AnnotationsVisualization"))
    print("Creating dataset:", args.output_dir)

    class_names = []
    class_name_to_id = {}
    for i, line in enumerate(open(args.labels).readlines()):
        class_id = i - 1  # starts with -1
        class_name = line.strip()
        class_name_to_id[class_name] = class_id
        if class_id == -1:
            assert class_name == "__ignore__"
            continue
        elif class_id == 0:
            assert class_name == "_background_"
        class_names.append(class_name)
    class_names = tuple(class_names)
    print("class_names:", class_names)
    out_class_names_file = osp.join(args.output_dir, "class_names.txt")
    with open(out_class_names_file, "w") as f:
        f.writelines("\n".join(class_names))
    print("Saved class_names:", out_class_names_file)

    for filename in glob.glob(osp.join(args.input_dir, "*.json")):
        print("Generating dataset from:", filename)

        label_file = labelme.LabelFile(filename=filename)

        base = osp.splitext(osp.basename(filename))[0]
        out_img_file = osp.join(args.output_dir, "JPEGImages", base + ".jpg")
        out_xml_file = osp.join(args.output_dir, "Annotations", base + ".xml")
        if not args.noviz:
            out_viz_file = osp.join(
                args.output_dir, "AnnotationsVisualization", base + ".jpg"
            )

        img = labelme.utils.img_data_to_arr(label_file.imageData)
        imgviz.io.imsave(out_img_file, img)

        maker = lxml.builder.ElementMaker()
        xml = maker.annotation(
            maker.folder(),
            maker.filename(base + ".jpg"),
            maker.database(),  # e.g., The VOC2007 Database
            maker.annotation(),  # e.g., Pascal VOC2007
            maker.image(),  # e.g., flickr
            maker.size(
                maker.height(str(img.shape[0])),
                maker.width(str(img.shape[1])),
                maker.depth(str(img.shape[2])),
            ),
            maker.segmented(),
        )

        bboxes = []
        labels = []
        for shape in label_file.shapes:
            # if shape["shape_type"] != "rectangle":
            #     print(
            #         "Skipping shape: label={label}, "
            #         "shape_type={shape_type}".format(**shape)
            #     )
            #     continue
            if shape["shape_type"] == "rectangle":
                class_name = shape["label"]
                class_id = class_names.index(class_name)

                (xmin, ymin), (xmax, ymax) = shape["points"]
                # swap if min is larger than max.
                xmin, xmax = sorted([xmin, xmax])
                ymin, ymax = sorted([ymin, ymax])

                bboxes.append([ymin, xmin, ymax, xmax])
                labels.append(class_id)

                xml.append(
                    maker.object(
                        maker.name(shape["label"]),
                        maker.pose(),
                        maker.truncated(),
                        maker.difficult(),
                        maker.bndbox(
                            maker.xmin(str(xmin)),
                            maker.ymin(str(ymin)),
                            maker.xmax(str(xmax)),
                            maker.ymax(str(ymax)),
                        ),
                    )
                )
            elif shape["shape_type"] == "point":
                class_name = shape["label"]
                class_id = class_names.index(class_name)
                # print(shape["points"])
                [[x,y]]= shape["points"]
                xml.append(
                    maker.object(
                        maker.name(shape["label"]),
                        maker.pose(),
                        maker.truncated(),
                        maker.difficult(),
                        maker.point(
                            maker.x(str(x)),
                            maker.y(str(y)),
                        ),
                    )
                )
            else:
                continue

        if not args.noviz:
            captions = [class_names[label] for label in labels]
            viz = imgviz.instances2rgb(
                image=img,
                labels=labels,
                bboxes=bboxes,
                captions=captions,
                font_size=15,
            )
            imgviz.io.imsave(out_viz_file, viz)

        with open(out_xml_file, "wb") as f:
            f.write(lxml.etree.tostring(xml, pretty_print=True))


if __name__ == "__main__":
    main()

生成如下的xml文件：

<annotation>
  <folder/>
  <filename>WIN_20231224_10_14_05_Pro.jpg</filename>
  <database/>
  <annotation/>
  <image/>
  <size>
    <height>1440</height>
    <width>2560</width>
    <depth>3</depth>
  </size>
  <segmented/>
  <object>
    <name>face</name>
    <pose/>
    <truncated/>
    <difficult/>
    <bndbox>
      <xmin>1222.1052631578948</xmin>
      <ymin>447.4436090225564</ymin>
      <xmax>1677.7443609022555</xmax>
      <ymax>1112.1052631578948</ymax>
    </bndbox>
  </object>
  <object>
    <name>le</name>
    <pose/>
    <truncated/>
    <difficult/>
    <point>
      <x>1383.7593984962405</x>
      <y>757.9699248120301</y>
    </point>
  </object>
  <object>
    <name>re</name>
    <pose/>
    <truncated/>
    <difficult/>
    <point>
      <x>1558.1954887218044</x>
      <y>758.7218045112782</y>
    </point>
  </object>
  <object>
    <name>no</name>
    <pose/>
    <truncated/>
    <difficult/>
    <point>
      <x>1477.7443609022555</x>
      <y>864.7368421052631</y>
    </point>
  </object>
  <object>
    <name>lm</name>
    <pose/>
    <truncated/>
    <difficult/>
    <point>
      <x>1400.3007518796992</x>
      <y>979.7744360902255</y>
    </point>
  </object>
  <object>
    <name>rm</name>
    <pose/>
    <truncated/>
    <difficult/>
    <point>
      <x>1540.9022556390976</x>
      <y>979.7744360902255</y>
    </point>
  </object>
</annotation>

3. xml 转换成 txt

借助人工智能写出代码框架，再进行调整，实现功能

import xml.etree.ElementTree as ET
import os

def operate(dir_path,file,result_path):
    file_name = os.path.join(dir_path, file)
    # 解析XML文件
    tree = ET.parse(file_name)
    root = tree.getroot()

    # 获取图片名称
    filename = root.find('filename').text

    # 遍历XML数据并转换为txt格式
    # fff = dir_path.replace("Annotations","JPEGImages") + "/"
    # txt_data = f"# {fff}{filename}\n"
    txt_data = f"# {filename}\n"

    for obj in root.findall('object'):
        name = obj.find('name').text
        if name == 'face':
            bndbox = obj.find('bndbox')
            xmin = float(bndbox.find('xmin').text)
            ymin = float(bndbox.find('ymin').text)
            xmax = float(bndbox.find('xmax').text)
            ymax = float(bndbox.find('ymax').text)
            txt_data += f"{int(xmin)} {int(ymin)} {int(xmax-xmin)} {int(ymax-ymin)} "

    for obj in root.findall('object'):
        name = obj.find('name').text
        if name in ["le"]:
            point = obj.find('point')
            x = float(point.find('x').text)
            y = float(point.find('y').text)
            txt_data += f"{int(x)} {int(y)} "

            txt_data += "0.0 "

    for obj in root.findall('object'):
        name = obj.find('name').text
        if name in ["re"]:
            point = obj.find('point')
            x = float(point.find('x').text)
            y = float(point.find('y').text)
            txt_data += f"{int(x)} {int(y)} "

            txt_data += "0.0 "

    for obj in root.findall('object'):
        name = obj.find('name').text
        if name in ["no"]:
            point = obj.find('point')
            x = float(point.find('x').text)
            y = float(point.find('y').text)
            txt_data += f"{int(x)} {int(y)} "

            txt_data += "0.0 "


    for obj in root.findall('object'):
        name = obj.find('name').text
        if name in ["lm"]:
            point = obj.find('point')
            x = float(point.find('x').text)
            y = float(point.find('y').text)
            txt_data += f"{int(x)} {int(y)} "

            txt_data += "0.0 "


    for obj in root.findall('object'):
        name = obj.find('name').text
        if name in ["rm"]:
            point = obj.find('point')
            x = float(point.find('x').text)
            y = float(point.find('y').text)
            txt_data += f"{int(x)} {int(y)} "

            txt_data += "0.0 "

    # 将转换后的txt数据写入文件
    with open(result_path, "a+",encoding="utf-8") as file:
        file.write(txt_data)
        file.write("\n")

dir_path = "wider/Annotations"
result_file_path = "result.txt"

dirs=os.listdir(dir_path)
for file in dirs:
    operate(dir_path,file,result_file_path)

得到如下内容的 .txt 文件

# WIN_20231224_10_14_05_Pro.jpg
1222 447 455 664 1383 757 0.0 1558 758 0.0 1477 864 0.0 1400 979 0.0 1540 979 0.0 
# WIN_20231224_10_14_06_Pro.jpg
1221 447 460 674 1386 762 0.0 1563 762 0.0 1479 866 0.0 1405 985 0.0 1551 988 0.0 
# WIN_20231224_10_14_07_Pro.jpg
1214 445 494 673 1497 774 0.0 1657 753 0.0 1617 875 0.0 1522 1005 0.0 1651 985 0.0 
# WIN_20231224_10_14_08_Pro.jpg
1203 451 475 684 1296 774 0.0 1471 778 0.0 1379 884 0.0 1331 1002 0.0 1466 1009 0.0 
# WIN_20231224_10_14_09_Pro.jpg
1221 436 447 642 1350 699 0.0 1522 691 0.0 1442 805 0.0 1378 944 0.0 1522 938 0.0 
# WIN_20231224_10_14_10_Pro.jpg
1216 448 464 712 1376 841 0.0 1552 839 0.0 1468 971 0.0 1398 1055 0.0 1531 1051 0.0 
# WIN_20231224_10_14_11_Pro.jpg
1201 446 466 665 1303 766 0.0 1493 761 0.0 1386 869 0.0 1331 988 0.0 1473 983 0.0 
# WIN_20231224_10_14_12_Pro (2).jpg
1295 459 519 692 1681 810 0.0 1794 810 0.0 1783 927 0.0 1651 1037 0.0 1745 1029 0.0 
# WIN_20231224_10_14_12_Pro.jpg
1216 454 469 672 1431 791 0.0 1603 778 0.0 1558 900 0.0 1469 1017 0.0 1601 1000 0.0