cvat_export_scripts

This repository contains a collection of custom cvat export scripts. The scripts are intended with the use for cvat.
NOTE: This repository is NOT by/from the developers of cvat! If you need cvat support, please go to the original github repository!

The purpose of this script is to use the cli provided in cvat to export the datasets in a more adequate folder-structure.

Note: These scripts are only compatible for version 2.1 of cvat!

MOT 1.1 The folder structure is split into train/test/validation. Each task is considered as a sequence i.e a video. That means each task is exported as a folder into train/test/validation accordingly.

The .seqinfo is created as well, altough not filled completly yet. You'll have to insert your own code. The images are stored in a subfolder called img1, the groundtruth in a subfolder gt.

YOLO 1.1 The export follows convention of Ultralytics custom dataset

Note: The generated .yaml file is NOT correct per default. You'll have to add the possible classifications afterwards.

Installation

You'll need to link to your cli instance of cvat. For this clone the repository of cvat with the same version as your cvat server instance.

The you have to adapt the python-variable in each script:

CVAT_CODE_LOCATION="/path/to/cvat/utils/cli"

Usage

If you want export all datasets from your cvat server in the MOT 1.1 format:

python export_mot.py --output <path/to/my/location>

Note: In the output folder there is going to be created a folder for each project in the cvat server!

If you want export all datasets from your cvat server in the YOLO 1.1 format:

python export_yolo.py --output <path/to/my/location>

Note: In the output folder there is going to be created a folder for each project in the cvat server!
o

Requirements

You need to clone the original cvat repository

pip install tqdm
pip install pexpect