When you use Yolo-model, you might create annotation labels with Yolo-mark.
For example,

• obj.names - example of list with object names
• train.txt - example with list of image filenames for training Yolo model
• train/ - example of folder that contain images and labels

• *.jpg : example of list of image
• *.txt : example of list of label

But, when you want to use another model(ex. efficientdet), you need another annotation format! 😥

• Oct 13th, 2021 - We could support not only Yolo-mark outputs, but also OpenLabeling outputs!
  Also, We could make segmentation mask polygons information in json file.
  Thanks to @NauchtanRobotics!

• numpy
• OpenCV

You can make same environment with anaconda environment.

• conda create -n Yolo-to-COCO python=3.8
• conda activate Yolo-to-COCO
• pip install numpy
• pip install opencv

Just clone this repository.

• git clone https://github.com/Taeyoung96/Yolo-to-COCO-format-converter.git
• cd Yolo-to-COCO-format-converter

When you have your own Yolo annotation format, just change a little bit!

In main.py, there is a code that declare the classes. You will change this with your obj.names.

Next, follow step 2 if you have your annotations in separate text files, one for each image. Alternatively, follow step 3 if you wish to work from YOLO annotations which are concatenated into a single file.

Use this approach if your training data file structure looks like this:

    dataset_root_dir/
        Photo_00001.jpg
        Photo_00001.txt
        Photo_00002.jpg
        Photo_00003.txt

You don't need to specify yolo-subdir argument.

• python main.py --path <Absolute path to dataset_root_dir> --output <Name of the json file>
• (For example)python main.py --path /home/taeyoungkim/Desktop/Yolo-to-COCO-format-converter/tutorial/ --output train

Use this approach if your annotations are in nested a level below the image files like this:

    dataset_root_dir/
        YOLO_darknet/
            Photo_00001.txt
            Photo_00002.txt
        Photo_00001.jpg
        Photo_00002.jpg

Command to use:

• python main.py --yolo-subdir --path <Absolute path to dataset_root_dir> --output <Name of the json file>
• python main.py --yolo-subdir --box2seg --path <Absolute path to dataset_root_dir> --output <Name of the json file>

The arg --box2seg initializes segmentation mask polygons that have box shapes. This is useful for when changing your modeling from object detection to image segmentation. These masks can then be reshaped using software such as the interface provided by makesense.ai

Make sure that it points to the absolute path to the folder where the image and text files are located.
You can easily change the path with Text Editor(Ubuntu 18.04) or NotePad (Window 10).

If you want to quickly create a train.txt file in Ubuntu, you can use path_replacer.py.

Works with 2 simple arguments.

• path_image_folder: File path where the images are located.
• path_txt: File path of the 'txt' file you want to create.

When you want to use

• python path_replacer.py --path_image_folder <File path where the images are located> --path_txt <File path of the 'txt' file you want to create>
• (For example)python path_replacer.py --path_image_folder /home/taeyoungkim/Desktop/Yolo-to-COCO-format-converter/tutorial/train --path_txt /home/taeyoungkim/Desktop/Yolo-to-COCO-format-converter/tutorial/train.txt

You need to provide 2 argments(essential) & 3 argments(optional).
essential

• path : Absolute path of train.txt
• output : Name of the json file

optional

• yolo-subdir : If your annotation label have OpenLabeling output.
• box2seg : If you want to make segmentation mask polygons that have box shapes.
• debug : If you want to check the bounding boxes or annotation information.

When you want to make json file,

• python main.py --path <Absolute Path of train.txt> --output <Name of the json file>
• (For example)python main.py --path /home/taeyoungkim/Desktop/Yolo-to-COCO-format-converter/tutorial/train.txt --output train

Or when you want to check the bounding boxes,

• python main.py --path <Absolute Path of train.txt> --output <Name of the json file> --debug
• (For example)python main.py --path /home/taeyoungkim/Desktop/Yolo-to-COCO-format-converter/tutorial/train.txt --output train --debug

If you want to read json files more clearly, you should use JQ!

• JQ Manual
• (For example) cd output
• jq . train.json > train_jq.json

Result of Json file

On debug mode, you can check bounding boxes

On debug mode, you can check annotation information on terminal

• I created a repository by referring to chrise96/image-to-coco-json-converter.
• GeeJae Lee helped to make it.

Copyright (c) 2021 Tae Young Kim

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.