vrdelc / COCO-dataset-from-labelme

Creating a COCO dataset from labelme's annotations

COCO dataset from labelme

Creating a COCO dataset from labelme's annotations

Requirements

Install Labelme

Know the path of the installation

Parameters

- -i: Name of the folder which contains the images and annotations
- -f: Name of the final folder with the COCO dataset
- -l: Path to the file with the labels of the dataset
- [-p: Path to the labelme installation (or path labelme doing: git clone https://github.com/wkentaro/labelme.git)] (Not now)
- -y: Year of the dataset. By default, it is 2019
- -t: Percetage of training. By default, it is 70

Structure of the final folder

- annotations
    - instances_train2019.json
    - instances_val2019.json
- train2019
    - image_a.jpeg
    - image_b.jpeg
    - ...
    - image_z.jpeg
- val2019
    - image_1.jpeg
    - image_2.jpeg
    - ...
    - image_15.jpeg