labelme标注图片，并生成dataset数据集方案【本地测试成功】

1.环境：win10+annaconda+labelme==3.16.2，python3.x

2.替换json_to_dataset.py文件，新内容如下：

1. import argparse
2. import json
3. import os
4. import os.path as osp
5. import warnings
7. import PIL.Image
8. import yaml
10. from labelme import utils
11. import base64
13. def main():
14.     warnings.warn("This script is aimed to demonstrate how to convert the\n"
15.                   "JSON file to a single image dataset, and not to handle\n"
16.                   "multiple JSON files to generate a real-use dataset.")
17.     parser = argparse.ArgumentParser()
18.     parser.add_argument('json_file')
19.     parser.add_argument('-o', '--out', default=None)
20.     args = parser.parse_args()
22.     json_file = args.json_file
23.     if args.out is None:
24.         out_dir = osp.basename(json_file).replace('.', '_')
25.         out_dir = osp.join(osp.dirname(json_file), out_dir)
26.     else:
27.         out_dir = args.out
28.     if not osp.exists(out_dir):
29.         os.mkdir(out_dir)
31.     count = os.listdir(json_file)
32.     for i in range(0, len(count)):
33.         path = os.path.join(json_file, count[i])
34.         if os.path.isfile(path):
35.             data = json.load(open(path))
36.             
37.             if data['imageData']:
38.                 imageData = data['imageData']
39.             else:
40.                 imagePath = os.path.join(os.path.dirname(path), data['imagePath'])
41.                 with open(imagePath, 'rb') as f:
42.                     imageData = f.read()
43.                     imageData = base64.b64encode(imageData).decode('utf-8')
44.             img = utils.img_b64_to_arr(imageData)
45.             label_name_to_value = {'_background_': 0}
46.             for shape in data['shapes']:
47.                 label_name = shape['label']
48.                 if label_name in label_name_to_value:
49.                     label_value = label_name_to_value[label_name]
50.                 else:
51.                     label_value = len(label_name_to_value)
52.                     label_name_to_value[label_name] = label_value
53.             
54.             # label_values must be dense
55.             label_values, label_names = [], []
56.             for ln, lv in sorted(label_name_to_value.items(), key=lambda x: x[1]):
57.                 label_values.append(lv)
58.                 label_names.append(ln)
59.             assert label_values == list(range(len(label_values)))
60.             
61.             lbl = utils.shapes_to_label(img.shape, data['shapes'], label_name_to_value)
62.             
63.             captions = ['{}: {}'.format(lv, ln)
64.                 for ln, lv in label_name_to_value.items()]
65.             lbl_viz = utils.draw_label(lbl, img, captions)
66.             
67.             out_dir = osp.basename(count[i]).replace('.', '_')
68.             out_dir = osp.join(osp.dirname(count[i]), out_dir)
69.             if not osp.exists(out_dir):
70.                 os.mkdir(out_dir)
72.             PIL.Image.fromarray(img).save(osp.join(out_dir, 'img.png'))
73.             #PIL.Image.fromarray(lbl).save(osp.join(out_dir, 'label.png'))
74.             utils.lblsave(osp.join(out_dir, 'label.png'), lbl)
75.             PIL.Image.fromarray(lbl_viz).save(osp.join(out_dir, 'label_viz.png'))
77.             with open(osp.join(out_dir, 'label_names.txt'), 'w') as f:
78.                 for lbl_name in label_names:
79.                     f.write(lbl_name + '\n')
81.             warnings.warn('info.yaml is being replaced by label_names.txt')
82.             info = dict(label_names=label_names)
83.             with open(osp.join(out_dir, 'info.yaml'), 'w') as f:
84.                 yaml.safe_dump(info, f, default_flow_style=False)
86.             print('Saved to: %s' % out_dir)
87. if __name__ == '__main__':
88.     main()

3.Labelme标注生成的JSON文件统一放在一个新的文件夹里面，不能有其它格式的文件。

 

4. 然后在pycharm里面新建一个py脚本文件，注意命名一定不能是json.py，随便取一个其它名字，最后运行该文件，即可生成数据集。

5.也可以使用命令生成数据集

6.参考资料：

（1）实现labelme批量json_to_dataset方法

https://blog.csdn.net/yql_617540298/article/details/81110685 

（2）初次使用labelme遇到的坑labelme_json_to_dataset 文件json只生成4个文件缺少生成infoyaml文件

https://www.pythonheidong.com/blog/article/401177/70aa441c62f6b74e4538/ 

（3）labelme 转换json为dataset，无需修改py文件

https://www.jianshu.com/p/9c803e83cbfe