Implementation:Obss Sahi Coco2yolo

Knowledge Sources	Obss_Sahi
Domains	Dataset, Format_Conversion
Last Updated	2026-02-08 12:00 GMT

Overview

CLI script that converts COCO-formatted annotation datasets to YOLO format with configurable train/val split.

Description

The coco2yolo script provides a command-line interface for converting COCO JSON annotation files into YOLO-formatted label files. It loads the COCO dataset using Coco.from_coco_dict_or_path(), then exports to YOLO format via coco.export_as_yolo(). The output includes properly split train/val directories with auto-incrementing experiment names.

Usage

Run this script when you need to convert COCO-formatted datasets into YOLO format for training with YOLOv5, Ultralytics, or other YOLO-family detectors.

Code Reference

Source Location

Repository: Obss_Sahi
File: sahi/scripts/coco2yolo.py
Lines: 1-48

Signature

def main(
    image_dir: str,
    dataset_json_path: str,
    train_split: int | float = 0.9,
    project: str = "runs/coco2yolo",
    name: str = "exp",
    seed: int = 1,
    disable_symlink=False,
) -> None:
    """
    Args:
        images_dir (str): directory for coco images
        dataset_json_path (str): file path for the coco json file to be converted
        train_split (float or int): set the training split ratio
        project (str): save results to project/name
        name (str): save results to project/name
        seed (int): fix the seed for reproducibility
        disable_symlink (bool): required in google colab env
    """

Import

from sahi.scripts.coco2yolo import main

I/O Contract

Inputs

Name	Type	Required	Description
image_dir	str	Yes	Directory containing COCO images
dataset_json_path	str	Yes	Path to COCO annotation JSON file
train_split	float or int	No (default 0.9)	Train/val split ratio
project	str	No (default "runs/coco2yolo")	Output project directory
name	str	No (default "exp")	Experiment name within project
seed	int	No (default 1)	Random seed for split reproducibility
disable_symlink	bool	No (default False)	Use file copy instead of symlinks (for Colab)

Outputs

Name	Type	Description
YOLO dataset	Directory	YOLO-formatted labels and images in project/name/

Usage Examples

CLI Usage

python -m sahi.scripts.coco2yolo \
    --image_dir /data/coco/images \
    --dataset_json_path /data/coco/annotations.json \
    --train_split 0.8 \
    --project runs/coco2yolo \
    --name my_dataset \
    --seed 42

Programmatic Usage

from sahi.scripts.coco2yolo import main

main(
    image_dir="/data/coco/images",
    dataset_json_path="/data/coco/annotations.json",
    train_split=0.9,
    project="runs/coco2yolo",
    name="exp",
    seed=1,
)

Related Pages

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment