darknet-to-supervisely

Conversion of Dataset from Darknet Format into Supervisely Format via Python

Table of Contents

• Introduction
• Usage
• Contributors
• Maintainers
• References
• License

Introduction

This repository shows a quick method to convert the dataset in the Darknet format into the JSON-based format required by Supervisely annotation tool in python language. Check it out!

Usage

Clone the repository to your local path:

git clone https://github.com/JinhangZhu/darknet-to-supervisely.git

Copy the convert.py file into your local folder where the darknet dataset is located. Open the terminal on Linux or command window in Windows and run command like:

python convert.py -o ./ego-hand -p ./new-project -d new-set -l left_hand -l right_hand

help:

• -o, --origin: The name of original data downloaded from Supervisely. type=str
• -p, --project: The name of the output dataset folder. type=str
• -d, --dataset: The name of the meta file of the data. type=str
• -l, --label: Whether to randomly split image set. action='append'

For example, I have a folder called: "ego-hand" in the current path, I want to create a dataset folder called "epichands" under the directory of project "epichhands", with labels: left_hand, right_hand. I run:

python convert.py -o ./ego-hand -p ./epichands -d epichands -l left_hand -l right_hand
Namespace(dataset='epichands', label=['left_hand', 'right_hand'], origin='./ego-hand', project='./epichands')
Images:   3%|██▌                                                                             | 415/12846 [00:13<06:29, 31.90it/s]

Results:

└───datasets                                                                   		
		├───ego-hand
		├───epichhands
				├───meta.json
				├───ann
					├───xxx.jpg.json
				├───img
					├───xxx.jpg

Contributors

Maintainers

Jinhang Zhu

References

• https://docs.supervise.ly/data-organization/import-export/supervisely-format

• https://docs.supervise.ly/data-organization/import-export/upload

License

• MIT