/oidv6-to-voc

Convert Open Image v6 detection dataset to VOC pasacal format XML. https://github.com/openimages/dataset

Primary LanguagePythonMIT LicenseMIT

OIDv6 To VOC

Convert bounding box datasets of Open Images Dataset v6 to VOC XML format.

Installation

pip3 install oidv6-to-voc

Usage

Once installed, you should be able to run it directly:

oidv6-to-voc -h

If your shell cannot find the command, try running it with:

python3 -m oidv6_to_voc -h

CLI options

To start converting, you need at least a part of the images, the class names metadata and at least one of the boxes annotation CSV file:

CSV files you need

oidv6-to-voc <annotation-file(s).csv>
             -d <class-names-file.csv> 
             --imgd <directory/to/your/images>
             --outd <your/output/diretory>

About the Dataset

The Open Images V6 Dataset contains 600 classes with 1900000+ images. The images are hosted on AWS, and the CSV files can be downloaded here.

To download it in full, you'll need 500+ GB of disk space. For downloading a part of the dataset only, I would recommend the DmitryRyumin/OIDv6 tool.

Credit

This repo is forked from AtriSaxena/OIDv4_to_VOC.