label-convert 0.0.2

A dataset format conversion tool for object detection and image segmentation, which supports mutual conversion between labelme, labelImg tools and YOLO, VOC, and COCO dataset formats.

Label Convert

A dataset format conversion tool for object detection and image segmentation, which supports mutual conversion between labelme, labelImg tools and YOLO, VOC, and COCO dataset formats.

Installation

pip install label_convert

Usage

labelImg label data → YOLOV5 format

Click to expand

• Convert the yolo data format marked by the labelImg library to YOLOV5 format data with one click.

• The labelImg label data directory structure is as follows (see dataset/labelImg_dataset for details):

  labelImg_dataset
  ├── classes.txt
  ├── images(13).jpg
  ├── images(13).txt
  ├── images(3).jpg
  ├── images(3).txt
  ├── images4.jpg
  ├── images4.txt
  ├── images5.jpg
  ├── images5.txt
  ├── images6.jpg
  ├── images7.jpg
  └── images7.txt
  

• Convert

  labelimg_to_yolov5 --src_dir dataset/labelImg_dataset \
                     --out_dir dataset/labelImg_dataset_output \
                     --val_ratio 0.2 \
                     --have_test true \
                     --test_ratio 0.2
  

  • --src_dir: the directory where labelImg is stored after labeling.
  • --out_dir: the location where the data is stored after conversion.
  • --val_ratio: the ratio of the generated validation set to the whole data, default is 0.2.
  • --have_test: whether to generate the test part of the data, the default is True.
  • --test_ratio: percentage of the whole data of the test data, default is 0.2.

• Converted directory structure (see dataset/labelImg_dataset_output for details):

  labelImg_dataset_output/
  ├── classes.txt
  ├── images
  │   ├── images(13).jpg
  │   ├── images(3).jpg
  │   ├── images4.jpg
  │   ├── images5.jpg
  │   └── images7.jpg
  ├── labels
  │   ├── images(13).txt
  │   ├── images(3).txt
  │   ├── images4.txt
  │   ├── images5.txt
  │   └── images7.txt
  ├── non_labels        # This is the catalog without the labeled images.
  │   └── images6.jpg
  ├── test.txt
  ├── train.txt
  └── val.txt
  

• You can further directly convert the dataset/labelImg_dataset_output directory to COCO

  yolov5_to_coco --data_dir dataset/labellImg_dataset_output
  

COCO format data → labelImg yolo format

Click to expand

• One-click conversion of COCO format data to labelImg labeled yolo format data.
• COCO format directory structure（see dataset/YOLOV5_COCO_format for details）：

  YOLOV5_COCO_format
  ├── annotations
  │   ├── instances_train2017.json
  │   └── instances_val2017.json
  ├── train2017
  │   ├── 000000000001.jpg
  │   └── 000000000002.jpg
  └── val2017
      └── 000000000001.jpg
  

• Convert

  coco_to_labelimg --data_dir dataset/YOLOV5_COCO_format
  

  • --data_dir: the directory where the COCO format dataset is located. Default is dataset/YOLOV5_COCO_format.
• Converted directory structure (see dataset/COCO_labelImg_format for details):

  COCO_labelImg_format
    ├── train
    │   ├── 000000000001.jpg
    │   ├── 000000000001.txt
    │   |-- 000000000002.jpg
    │   └── classes.txt
    └── val
        ├── 000000000001.jpg
        ├── 000000000001.txt
        └── classes.txt
  

• For the converted directory, you can directly use the labelImg library to open it directly and change the label. The specific commands are as follows:

  $ cd dataset/COCO_labelImg_format
  $ labelImg train train/classes.txt
  
  # or
  $ labelImg val val/classes.txt
  

YOLOV5 format data → COCO

Click to expand

• Some background images can be added to the training by directly placing them into the backgroud_images directory.

• The conversion program will automatically scan this directory and add it to the training set, allowing seamless integration with subsequent YOLOX training.

• YOLOV5 training format directory structure (see dataset/YOLOV5 for details).

  YOLOV5
  ├── classes.txt
  ├── background_images  # usually images that are easily confused with the object to be detected
  │   └── bg1.jpeg
  ├── images
  │   ├── images(13).jpg
  │   └── images(3).jpg
  ├── labels
  │   ├── images(13).txt
  │   └── images(3).txt
  ├── train.txt
  └── val.txt
  

• The image paths in train.txt and val.txt can be either:

  • Path relative to root directory:

    dataset/YOLOV5/images/images(3).jpg
    

  • Path relative to dataset/YOLOV5:

    images/images(3).jpg
    

• Convert

  yolov5_to_coco --data_dir dataset/YOLOV5 --mode_list train,val
  

  • --data_dir: the directory where the collated dataset is located
  • --mode_list: specify the generated json, provided that there is a corresponding txt file, which can be specified separately. (e.g. train,val,test)

• The structure of the converted directory (see dataset/YOLOV5_COCO_format for details)

  YOLOV5_COCO_format
  ├── annotations
  │   ├── instances_train2017.json
  │   └── instances_val2017.json
  ├── train2017
  │   ├── 000000000001.jpg
  │   └── 000000000002.jpg  # This is the background image.
  └── val2017
      └── 000000000001.jpg
  

YOLOV5 YAML description file → COCO

Click to expand

• The YOLOV5 yaml data file needs to contain.

  YOLOV5_yaml
  ├── images
  │   ├── train
  │   │   ├── images(13).jpg
  │   │   └── images(3).jpg
  │   └── val
  │       ├── images(13).jpg
  │       └── images(3).jpg
  ├── labels
  │   ├── train
  │   │   ├── images(13).txt
  │   │   └── images(3).txt
  │   └── val
  │       ├── images(13).txt
  │       └── images(3).txt
  └── sample.yaml
  

• Convert

  yolov5_yaml_to_coco --yaml_path dataset/YOLOV5_yaml/sample.yaml
  

darknet format data → COCO

• Darknet training data directory structure (see dataset/darknet for details).

  darknet
  ├── class.names
  ├── gen_config.data
  ├── gen_train.txt
  ├── gen_valid.txt
  └── images
      ├── train
      └── valid
  

• Convert

  darknet_to_coco --data_path dataset/darknet/gen_config.data
  

Visualize images in COCO format

Click to expand

coco_visual --vis_num 1 \
            --json_path dataset/YOLOV5_COCO_format/annotations/instances_train2017.json \
            --img_dir dataset/YOLOV5_COCO_format/train2017

• --vis_num: specify the index of the image to be viewed
• --json_path: path to the json file of the image to view
• --img_dir: view the directory where the image is located

Object Instance demo of COCO

Click to expand

{
    "info": {
      "year": 2022,
      "version": "1.0",
      "description": "For object detection",
      "date_created": "2022"
    },
    "licenses":  [{
        "id": 1,
        "name": "Apache License v2.0",
        "url": "https://github.com/RapidAI/YOLO2COCO/LICENSE"
    }],
    "images": [{
        "date_captured": "2022",
        "file_name": "000000000001.jpg",
        "id": 1,
        "height": 224,
        "width": 224
    }, {
        "date_captured": "2022",
        "file_name": "000000000002.jpg",
        "id": 2,
        "height": 424,
        "width": 550
    }],
    "annotations": [{
        "segmentation": [[18.00, 2.99, 105.00, 2.99, 105.00, 89.00, 18.00, 89.00]],
        "area": 7482.011,
        "iscrowd": 0,
        "image_id": 1,  // Corresponding to the ID in images
        "bbox": [18.00, 2.99, 87.00, 86.00],  // [x, y, w, h], (x,y) is the left top point of the box. w,h is the width and height of the box.
        "category_id": 1,  // Corresponding to the ID in categories.
        "id": 1  // Number that uniquely distinguishes different dimension instances
    }, {
        "segmentation": [
            [126.99, 3.99, 210.99, 3.99, 210.99, 88.99, 126.99, 88.99]
        ],
        "area": 7139.994,
        "iscrowd": 0,
        "image_id": 1,
        "bbox": [126.99, 3.99, 84.0, 84.99],
        "category_id": 1,
        "id": 2
    }],
    "categories": [{
        "supercategory": "stamp",
        "id": 1,
        "name": "stamp"
    }]
}

Related information

• MSCOCO Data Annotation Details

See details for LabelConvert