coco-to-yolo

Simple command line tool to convert COCO object detection datasets to YOLO format.

Usage

1. Install via pip pip install coco-to-yolo
2. Convert COCO dataset to ultralytics YOLO format using coco_to_yolo <<coco_dir>> <<output_dir>>

By default the script assumes the coco dataset to be structured as follows:

<<coco_dir>>
├── annotations
│   └── annotations.json     # Exactly ONE annotation file in COCO json format
└── images                   # Arbitary number of images (matching the file names 
    ├── image1.jpeg          # in the annotations json file)
    ├── image2.jpeg         
    └── ...

Split generation

By default the script will split 10% of the data into a test split and not generate a validation split. The ratio of splitted test and validation data can be adapted by specifying the --test_ratio and --val_ratio arguments, e.g.

Example usage

coco_to_yolo /home/COCO_ds /home/COCO_ds --test_ratio 0.15 --val_ratio 0.1

will convert the dataset in the /home/COCO_ds to the format required by YOLO, split 15% of the data for the testing, 10% for validation, and store the resulting dataset in /home/COCO_ds

Contributions

Feel free to suggest extensions and point out mistakes by creating an issue or sending me a pull request.