YOLO segmentation data set production: use the Labelme tool to create a segmentation data set (.json) and convert it to the format of the YOLO data set (.txt)

1. Labelme production data set

• Open labelme, label the data, and generate the file directory as follows:
  • data
    • a.jpg
    • a.json
    • b.jpg
    • b.json
    • …

2. Convert the data in labelme format to coco format (the converted coco format is still a json file)

• Code download: instance_segmentation

• Revise:

  • 78 lines found

    class_name_to_id = {}
    for i, line in enumerate(open(args.labels).readlines()):
        # class_id = i-1  # starts with -1
        class_id = i ## 没有背景，所以索引地址从0开始
        class_name = line.strip()
        if class_id == -1:
            assert class_name == "__ignore__"
            continue
    

  • Create labels.txt (in the same file directory as the data folder)

    • data
      • a.jpg
      • a.json
      • b.jpg
      • b.json
      • …
    • labels.txt

    labels.txt content:

    ```dotnetcli
    __ignore__
    class 0
    class 1
    ....
    
    ```
    

• Start converting json format

  python labelme2coco.py ..path/data ..path/data_dataset_coco --labels labels.txt
  

3. Convert coco format to YOLO format

• File download: COCO2YOLO

• Revise

  • Modification 1: Find 306 lines

    ## with open((fn / f).with_suffix('.txt'), 'a') as file:
    with open((fn / f[11:]).with_suffix('.txt'), 'a') as file:
    

  • Modification 2: Find the main function: modify the json path to your file path

    if source == 'COCO':
        convert_coco_json('..path/data_dataset_coco',  # directory with *.json
                          use_segments=True,
                          cls91to80=True)
    

• Start converting json format

  python general_json2yolo.py
  

• The generated yolo annotation file is in: ./new_dir/labels/annotations/