devbox
盐析白兔
这个屌丝很懒，什么也没留下！
关注作者
热门标签
热门文章
当前位置:   article > 正文

将labelme标注的人体姿态Json文件转成训练Yolov8-Pose的txt格式_labelme骨骼点json转yolo

作者：盐析白兔 | 2024-04-23 14:15:50

赞

踩

labelme骨骼点json转yolo

最近在训练Yolov8-Pose时遇到一个问题，就是如何将自己使用labelme标注的Json文件转化成可用于Yolov8-Pose训练的txt文件。

具体代码有以下：

1、labelme2coco.py 将自己标注的人体姿态信息json文件格式合并转换成CoCo格式

  1.  
  2.  
  3. import os
  4. import json
  5. import numpy as np
  6. import glob
  7. import shutil
  8. np.random.seed(41)
  9. import cv2
  10.  
  11. #0为背景
  12. classname_to_id = {"person": 1}
  13.  
  14. class Lableme2CoCo:
  15.  
  16.     def __init__(self, splitDir=''):
  17.         self.images = []
  18.         self.annotations = []
  19.         self.categories = []
  20.         self.img_id = 0
  21.         self.ann_id = 0
  22.         self.splitDir = splitDir
  23.  
  24.     def save_coco_json(self, instance, save_path):
  25.         json.dump(instance, open(save_path, 'w', encoding='utf-8'), ensure_ascii=False, indent=1)  # indent=2 更加美观显示
  26.  
  27.     # 由json文件构建COCO
  28.     def to_coco(self, json_path_list):
  29.         self._init_categories()
  30.         for json_path in json_path_list:
  31.             # print(type(json_path))
  32.             obj = self.read_jsonfile(json_path)
  33.             self.images.append(self._image(obj, json_path))
  34.  
  35.             shapes = obj['shapes']
  36.             groupIds = []
  37.             for shape in shapes:
  38.                 groupId = shape['group_id']
  39.                 groupIds.append(groupId)
  40.             for i in set(groupIds):
  41.                 keyPoints = [0] * 51
  42.                 keyPointNum = 0
  43.                 bbox = []
  44.  
  45.                 for shape in shapes:
  46.                     if i != shape['group_id']:
  47.                         continue
  48.                     if shape['shape_type'] == "point":
  49.                         labelNum = int(shape['label'])
  50.                         keyPoints[labelNum * 3 + 0] = int(shape['points'][0][0] + 0.5)
  51.                         keyPoints[labelNum * 3 + 1] = int(shape['points'][0][1] + 0.5)
  52.                         keyPoints[labelNum * 3 + 2] = 2
  53.                         keyPointNum += 1
  54.                     if shape['shape_type'] == 'rectangle':
  55.                         x0, y0, x1, y1 = shape['points'][0][0], shape['points'][0][1], \
  56.                                          shape['points'][1][0], shape['points'][1][1]
  57.                         xmin = min(x0, x1)
  58.                         ymin = min(y0, y1)
  59.                         xmax = max(x0, x1)
  60.                         ymax = max(y0, y1)
  61.  
  62.                         bbox = [xmin, ymin, xmax - xmin, ymax - ymin]
  63.  
  64.                 annotation = self._annotation(bbox, keyPoints, keyPointNum)
  65.                 self.annotations.append(annotation)
  66.                 self.ann_id += 1
  67.             self.img_id += 1
  68.  
  69.             # for shape in shapes:
  70.             #     label = shape['label']
  71.             #     if label != 'person':
  72.             #         continue
  73.             #
  74.             #     annotation = self._annotation(shape)
  75.             #     self.annotations.append(annotation)
  76.             #     self.ann_id += 1
  77.             # self.img_id += 1
  78.         instance = {}
  79.         instance['info'] = 'spytensor created'
  80.         instance['license'] = ['license']
  81.         instance['images'] = self.images
  82.         instance['annotations'] = self.annotations
  83.         instance['categories'] = self.categories
  84.         return instance
  85.  
  86.     # 构建类别
  87.     def _init_categories(self):
  88.         for k, v in classname_to_id.items():
  89.             category = {}
  90.             category['id'] = v
  91.             category['name'] = k
  92.             self.categories.append(category)
  93.  
  94.     # 构建COCO的image字段
  95.     def _image(self, obj, jsonPath):
  96.         image = {}
  97.         # img_x = utils.img_b64_to_arr(obj['imageData'])
  98.         # h, w = img_x.shape[:-1]
  99.         jpgPath = jsonPath.replace('.json', '.jpg')
  100.         jpgData = cv2.imread(jpgPath)
  101.         h, w, _ = jpgData.shape
  102.  
  103.         image['height'] = h
  104.         image['width'] = w
  105.         image['id'] = self.img_id
  106.  
  107.         # image['file_name'] = os.path.basename(jsonPath).replace(".json", ".jpg")
  108.         image['file_name'] = jpgPath.split(self.splitDir)[-1].replace('\\', '/')
  109.  
  110.  
  111.         return image
  112.  
  113.     # 构建COCO的annotation字段
  114.     def _annotation(self, bbox, keyPoints, keyNum):
  115.         annotation = {}
  116.  
  117.         annotation['id'] = self.ann_id
  118.         annotation['image_id'] = self.img_id
  119.         annotation['category_id'] = 1
  120.         # annotation['segmentation'] = [np.asarray(points).flatten().tolist()]
  121.         annotation['segmentation'] = []
  122.         annotation['bbox'] = bbox
  123.         annotation['iscrowd'] = 0
  124.         annotation['area'] = bbox[2] * bbox[3]
  125.         annotation['keypoints'] = keyPoints
  126.         annotation['num_keypoints'] = keyNum
  127.  
  128.         return annotation
  129.  
  130.     # 读取json文件，返回一个json对象
  131.     def read_jsonfile(self, path):
  132.         with open(path, "r", encoding='utf-8') as f:
  133.             return json.load(f)
  134.  
  135.     # COCO的格式： [x1,y1,w,h] 对应COCO的bbox格式
  136.     def _get_box(self, points):
  137.         min_x = min_y = np.inf
  138.         max_x = max_y = 0
  139.         for x, y in points:
  140.             min_x = min(min_x, x)
  141.             min_y = min(min_y, y)
  142.             max_x = max(max_x, x)
  143.             max_y = max(max_y, y)
  144.         return [min_x, min_y, max_x - min_x, max_y - min_y]
  145.  
  146.  
  147. if __name__ == '__main__':
  148.  
  149.     labelme_path = r"G:\XRW\Data\selfjson"
  150.     print(labelme_path)
  151.     jsonName = labelme_path.split('\\')[-1]
  152.     saved_coco_path = r"G:\XRW\Data\mycoco"
  153.     print(saved_coco_path)
  154.     #####################################
  155.     # 这个一定要注意
  156.     # 为了方便合入coco数据， 定义截断文件的文件夹与文件名字
  157.     splitDirFlag = 'labelMePoint\\'
  158.     ######################################
  159.  
  160.     # 创建文件
  161.     if not os.path.exists("%s/annotations/"%saved_coco_path):
  162.         os.makedirs("%s/annotations/"%saved_coco_path)
  163.     json_list_path = glob.glob(os.path.join(labelme_path, '*.json'))
  164.  
  165.     train_path, val_path = json_list_path, ''
  166.     # print(train_path)
  167.     print("train_n:", len(train_path), 'val_n:', len(val_path))
  168.  
  169.     # 把训练集转化为COCO的json格式
  170.     l2c_train = Lableme2CoCo(splitDirFlag)
  171.     # print(train_path)
  172.     train_instance = l2c_train.to_coco(train_path)
  173.     l2c_train.save_coco_json(train_instance, '%s/annotations/%s.json'%(saved_coco_path, jsonName))
  174.  
  175.

labelme_path：自己标注的json文件路径

saved_coco_path：生成的CoCo格式保存位置

运行代码得到

2、将生成的CoCo格式的Json文件转换成Yolov8-Pose格式的txt文件

utils.py

  1. import glob
  2. import os
  3. import shutil
  4. from pathlib import Path
  5.  
  6. import numpy as np
  7. from PIL import ExifTags
  8. from tqdm import tqdm
  9.  
  10. # Parameters
  11. img_formats = ['bmp', 'jpg', 'jpeg', 'png', 'tif', 'tiff', 'dng']  # acceptable image suffixes
  12. vid_formats = ['mov', 'avi', 'mp4', 'mpg', 'mpeg', 'm4v', 'wmv', 'mkv']  # acceptable video suffixes
  13.  
  14. # Get orientation exif tag
  15. for orientation in ExifTags.TAGS.keys():
  16.     if ExifTags.TAGS[orientation] == 'Orientation':
  17.         break
  18.  
  19.  
  20. def exif_size(img):
  21.     # Returns exif-corrected PIL size
  22.     s = img.size  # (width, height)
  23.     try:
  24.         rotation = dict(img._getexif().items())[orientation]
  25.         if rotation in [6, 8]:  # rotation 270
  26.             s = (s[1], s[0])
  27.     except:
  28.         pass
  29.  
  30.     return s
  31.  
  32.  
  33. def split_rows_simple(file='../data/sm4/out.txt'):  # from utils import *; split_rows_simple()
  34.     # splits one textfile into 3 smaller ones based upon train, test, val ratios
  35.     with open(file) as f:
  36.         lines = f.readlines()
  37.  
  38.     s = Path(file).suffix
  39.     lines = sorted(list(filter(lambda x: len(x) > 0, lines)))
  40.     i, j, k = split_indices(lines, train=0.9, test=0.1, validate=0.0)
  41.     for k, v in {'train': i, 'test': j, 'val': k}.items():  # key, value pairs
  42.         if v.any():
  43.             new_file = file.replace(s, f'_{k}{s}')
  44.             with open(new_file, 'w') as f:
  45.                 f.writelines([lines[i] for i in v])
  46.  
  47.  
  48. def split_files(out_path, file_name, prefix_path=''):  # split training data
  49.     file_name = list(filter(lambda x: len(x) > 0, file_name))
  50.     file_name = sorted(file_name)
  51.     i, j, k = split_indices(file_name, train=0.9, test=0.1, validate=0.0)
  52.     datasets = {'train': i, 'test': j, 'val': k}
  53.     for key, item in datasets.items():
  54.         if item.any():
  55.             with open(f'{out_path}_{key}.txt', 'a') as file:
  56.                 for i in item:
  57.                     file.write('%s%s\n' % (prefix_path, file_name[i]))
  58.  
  59.  
  60. def split_indices(x, train=0.9, test=0.1, validate=0.0, shuffle=True):  # split training data
  61.     n = len(x)
  62.     v = np.arange(n)
  63.     if shuffle:
  64.         np.random.shuffle(v)
  65.  
  66.     i = round(n * train)  # train
  67.     j = round(n * test) + i  # test
  68.     k = round(n * validate) + j  # validate
  69.     return v[:i], v[i:j], v[j:k]  # return indices
  70.  
  71.  
  72. def make_dirs(dir='new_dir/'):
  73.     # Create folders
  74.     dir = Path(dir)
  75.     if dir.exists():
  76.         shutil.rmtree(dir)  # delete dir
  77.     for p in dir, dir / 'labels', dir / 'images':
  78.         p.mkdir(parents=True, exist_ok=True)  # make dir
  79.     return dir
  80.  
  81.  
  82. def write_data_data(fname='data.data', nc=80):
  83.     # write darknet *.data file
  84.     lines = ['classes = %g\n' % nc,
  85.              'train =../out/data_train.txt\n',
  86.              'valid =../out/data_test.txt\n',
  87.              'names =../out/data.names\n',
  88.              'backup = backup/\n',
  89.              'eval = coco\n']
  90.  
  91.     with open(fname, 'a') as f:
  92.         f.writelines(lines)
  93.  
  94.  
  95. def image_folder2file(folder='images/'):  # from utils import *; image_folder2file()
  96.     # write a txt file listing all imaged in folder
  97.     s = glob.glob(f'{folder}*.*')
  98.     with open(f'{folder[:-1]}.txt', 'w') as file:
  99.         for l in s:
  100.             file.write(l + '\n')  # write image list
  101.  
  102.  
  103. def add_coco_background(path='../data/sm4/', n=1000):  # from utils import *; add_coco_background()
  104.     # add coco background to sm4 in outb.txt
  105.     p = f'{path}background'
  106.     if os.path.exists(p):
  107.         shutil.rmtree(p)  # delete output folder
  108.     os.makedirs(p)  # make new output folder
  109.  
  110.     # copy images
  111.     for image in glob.glob('../coco/images/train2014/*.*')[:n]:
  112.         os.system(f'cp {image} {p}')
  113.  
  114.     # add to outb.txt and make train, test.txt files
  115.     f = f'{path}out.txt'
  116.     fb = f'{path}outb.txt'
  117.     os.system(f'cp {f} {fb}')
  118.     with open(fb, 'a') as file:
  119.         file.writelines(i + '\n' for i in glob.glob(f'{p}/*.*'))
  120.     split_rows_simple(file=fb)
  121.  
  122.  
  123. def create_single_class_dataset(path='../data/sm3'):  # from utils import *; create_single_class_dataset('../data/sm3/')
  124.     # creates a single-class version of an existing dataset
  125.     os.system(f'mkdir {path}_1cls')
  126.  
  127.  
  128. def flatten_recursive_folders(path='../../Downloads/data/sm4/'):  # from utils import *; flatten_recursive_folders()
  129.     # flattens nested folders in path/images and path/JSON into single folders
  130.     idir, jdir = f'{path}images/', f'{path}json/'
  131.     nidir, njdir = Path(f'{path}images_flat/'), Path(f'{path}json_flat/')
  132.     n = 0
  133.  
  134.     # Create output folders
  135.     for p in [nidir, njdir]:
  136.         if os.path.exists(p):
  137.             shutil.rmtree(p)  # delete output folder
  138.         os.makedirs(p)  # make new output folder
  139.  
  140.     for parent, dirs, files in os.walk(idir):
  141.         for f in tqdm(files, desc=parent):
  142.             f = Path(f)
  143.             stem, suffix = f.stem, f.suffix
  144.             if suffix.lower()[1:] in img_formats:
  145.                 n += 1
  146.                 stem_new = '%g_' % n + stem
  147.                 image_new = nidir / (stem_new + suffix)  # converts all formats to *.jpg
  148.                 json_new = njdir / f'{stem_new}.json'
  149.  
  150.                 image = parent / f
  151.                 json = Path(parent.replace('images', 'json')) / str(f).replace(suffix, '.json')
  152.  
  153.                 os.system("cp '%s' '%s'" % (json, json_new))
  154.                 os.system("cp '%s' '%s'" % (image, image_new))
  155.                 # cv2.imwrite(str(image_new), cv2.imread(str(image)))
  156.  
  157.     print('Flattening complete: %g jsons and images' % n)
  158.  
  159.  
  160. def coco91_to_coco80_class():  # converts 80-index (val2014) to 91-index (paper)
  161.     # https://tech.amikelive.com/node-718/what-object-categories-labels-are-in-coco-dataset/
  162.     x = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, None, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, None, 24, 25, None,
  163.          None, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, None, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50,
  164.          51, 52, 53, 54, 55, 56, 57, 58, 59, None, 60, None, None, 61, None, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72,
  165.          None, 73, 74, 75, 76, 77, 78, 79, None]
  166.     return x
slefjson2posetxt.py
  1.  
  2.  
  3. import json
  4.  
  5. import cv2
  6. import pandas as pd
  7. from PIL import Image
  8. from collections import defaultdict
  9.  
  10. from utils import *
  11.  
  12.  
  13.  
  14. def convert_coco_json(cocojsonpath, savepath,use_keypoints=False, cls91to80=True):
  15.     """Converts COCO dataset annotations to a format suitable for training YOLOv5 models.
  17.     Args:
  18.         labels_dir (str, optional): Path to directory containing COCO dataset annotation files.
  19.         use_segments (bool, optional): Whether to include segmentation masks in the output.
  20.         use_keypoints (bool, optional): Whether to include keypoint annotations in the output.
  21.         cls91to80 (bool, optional): Whether to map 91 COCO class IDs to the corresponding 80 COCO class IDs.
  23.     Raises:
  24.         FileNotFoundError: If the labels_dir path does not exist.
  26.     Example Usage:
  27.         convert_coco(labels_dir='../coco/annotations/', use_segments=True, use_keypoints=True, cls91to80=True)
  29.     Output:
  30.         Generates output files in the specified output directory.
  31.     """
  32.     # save_dir = make_dirs('yolo_labels')  # output directory
  33.     save_dir = make_dirs(savepath)  # output directory
  34.     coco80 = coco91_to_coco80_class()
  35.  
  36.     # Import json
  37.     for json_file in sorted(Path(cocojsonpath).resolve().glob('*.json')):
  38.         fn = Path(save_dir) / 'labels' / json_file.stem.replace('instances_', '')  # folder name
  39.         fn.mkdir(parents=True, exist_ok=True)
  40.         with open(json_file) as f:
  41.             data = json.load(f)
  42.  
  43.         # Create image dict
  44.         images = {f'{x["id"]:d}': x for x in data['images']}
  45.         # Create image-annotations dict
  46.         imgToAnns = defaultdict(list)
  47.         for ann in data['annotations']:
  48.             imgToAnns[ann['image_id']].append(ann)
  49.  
  50.         # Write labels file
  51.         for img_id, anns in tqdm(imgToAnns.items(), desc=f'Annotations {json_file}'):
  52.             img = images[f'{img_id:d}']
  53.             h, w, f = img['height'], img['width'], img['file_name']
  54.  
  55.             bboxes = []
  56.             segments = []
  57.             keypoints = []
  58.             for ann in anns:
  59.                 if ann['iscrowd']:
  60.                     continue
  61.                 # The COCO box format is [top left x, top left y, width, height]
  62.                 box = np.array(ann['bbox'], dtype=np.float64)
  63.                 box[:2] += box[2:] / 2  # xy top-left corner to center
  64.                 box[[0, 2]] /= w  # normalize x
  65.                 box[[1, 3]] /= h  # normalize y
  66.                 if box[2] <= 0 or box[3] <= 0:  # if w <= 0 and h <= 0
  67.                     continue
  68.  
  69.                 cls = coco80[ann['category_id'] - 1] if cls91to80 else ann['category_id'] - 1  # class
  70.                 box = [cls] + box.tolist()
  71.                 if box not in bboxes:
  72.                     bboxes.append(box)
  73.                 if use_keypoints and ann.get('keypoints') is not None:
  74.                     k = (np.array(ann['keypoints']).reshape(-1, 3) / np.array([w, h, 1])).reshape(-1).tolist()
  75.                     k = box + k
  76.                     keypoints.append(k)
  77.  
  78.             # Write
  79.             fname = f.split('/')[-1]
  80.             # with open((fn / f).with_suffix('.txt'), 'a') as file:
  81.             with open((fn / fname).with_suffix('.txt'), 'a') as file:
  82.                 for i in range(len(bboxes)):
  83.                     if use_keypoints:
  84.                         line = *(keypoints[i]),  # cls, box, keypoints
  85.                     file.write(('%g ' * len(line)).rstrip() % line + '\n')
  86. if __name__ == '__main__':
  87.     source = 'COCO'
  88.     cocojsonpath = r'G:\XRW\Data\mycoco\annotations'
  89.     savepath = r'G:\XRW\Data\myposedata'
  90.     if source == 'COCO':
  91.         convert_coco_json(cocojsonpath,  # directory with *.json
  92.                           savepath,
  93.                           use_keypoints=True,
  94.                           cls91to80=True)

运行代码得到：

<class-index>是对象的类的索引，<x> <y> <width> <height>是边界框的坐标，<px1> <py1> <px2> <py2> ... <pxn> <pyn>是关键点的像素坐标。坐标由空格分隔。

3、检查生成的txt是否准确

PoseVisualization.py：将txt的信息可视化在图片上进行验证。
  1.  
  2.  
  3. import cv2
  4. imgpath = r'G:\XRW\Data\selfjson\five_22101205_000930.jpg'
  5. txtpath = r'G:\XRW\Data\myposedata\labels\selfjson\five_22101205_000930.txt'
  6.  
  7. f = open(txtpath,'r')
  8. lines = f.readlines()
  9. img = cv2.imread(imgpath)
  10. h, w, c = img.shape
  11. colors = [[255, 128, 0], [255, 153, 51], [255, 178, 102], [230, 230, 0], [255, 153, 255],
  12.                                       [153, 204, 255], [255, 102, 255], [255, 51, 255], [102, 178, 255], [51, 153, 255],
  13.                                       [255, 153, 153], [255, 102, 102], [255, 51, 51], [153, 255, 153], [102, 255, 102],
  14.                                       [51, 255, 51], [0, 255, 0], [0, 0, 255], [255, 0, 0], [255, 255, 255]]
  15. for line in lines:
  16.     print(line)
  17.     l = line.split(' ')
  18.     print(len(l))
  19.     cx = float(l[1]) * w
  20.     cy = float(l[2]) * h
  21.     weight = float(l[3]) * w
  22.     height = float(l[4]) * h
  23.     xmin = cx - weight/2
  24.     ymin = cy - height/2
  25.     xmax = cx + weight/2
  26.     ymax = cy + height/2
  27.     print((xmin,ymin),(xmax,ymax))
  28.     cv2.rectangle(img,(int(xmin),int(ymin)),(int(xmax),int(ymax)),(0,255,0),2)
  29.     kpts = []
  30.  
  31.     for i in range(17):
  32.         x = float(l[5:][3*i]) * w
  33.         y = float(l[5:][3*i+1]) * h
  34.         s = int(l[5:][3*i+2])
  35.         print(x,y,s)
  36.         if s != 0:
  37.             cv2.circle(img,(int(x),int(y)),1,colors[i],2)
  38.         kpts.append([int(x),int(y),int(s)])
  39.     print(kpts)
  40.     kpt_line = [[16, 14], [14, 12], [17, 15], [15, 13], [12, 13], [6, 12], [7, 13], [6, 7], [6, 8], [7, 9],
  41.                 [8, 10], [9, 11], [2, 3], [1, 2], [1, 3], [2, 4], [3, 5], [4, 6], [5, 7]]
  42.     for j in range(len(kpt_line)):
  43.         m,n = kpt_line[j][0],kpt_line[j][1]
  44.         if kpts[m-1][2] !=0 and kpts[n-1][2] !=0:
  45.             cv2.line(img,(kpts[m-1][0],kpts[m-1][1]),(kpts[n-1][0],kpts[n-1][1]),colors[j],2)
  46.  
  47. img = cv2.resize(img, None, fx=0.5, fy=0.5)
  48. cv2.imshow('1',img)
  49. cv2.waitKey(0)

这样就将自己的Json格式转成训练Yolov8-Pose姿态的txt格式了。

4、将图片copy到对应路径中

以上步骤完成后只生成了txt，需要再将对应的图片copy到对应路径中。

pickImg.py

  1.  
  2. import glob
  3. import os
  4. import shutil
  5. imgpath = r'G:\XRW\Data\selfjson'
  6. txtpath = r'G:\XRW\Data\myposedata\labels\selfjson'
  7. savepath = r'G:\XRW\Data\myposedata\images\selfjson'
  8. os.makedirs(savepath,exist_ok=True)
  9.  
  10. imglist = glob.glob(os.path.join(imgpath ,'*.jpg'))
  11. # print(imglist)
  12. txtlist = glob.glob(os.path.join(txtpath ,'*.txt'))
  13. # print(txtlist)
  14. for img in imglist:
  15.     name = txtpath + '\\' +img.split('\\')[-1].split('.')[0 ] +'.txt'
  16.     print(name)
  17.     if name in txtlist:
  18.         shutil.copy(img ,savepath)
  • imgpath CoCo数据集图片路径
  • txtpath 生成的txt路径
  • savepath 保存图片的路径

这样就将自己标注的数据集转换成Yolov8-Pose格式的txt了。

声明：本文内容由网友自发贡献，不代表【wpsshop博客】立场，版权归原作者所有，本站不承担相应法律责任。如您发现有侵权的内容，请联系我们。转载请注明出处：https://www.wpsshop.cn/w/盐析白兔/article/detail/474298
推荐阅读
相关标签

Copyright © 2003-2013 www.wpsshop.cn 版权所有，并保留所有权利。

  

闽ICP备14008679号