VME: A Satellite Imagery Dataset and Benchmark for Detecting Vehicles in the Middle East and Beyond

This repository has VME dataset (images and annotations files). Also, it has the script for constructing CDSI dataset.

VME is a satellite imagery dataset built for vehicle detection in the Middle East. VME images (satellite_images folder) are under CC BY-NC-ND 4.0 license, whereas the rest of folders (annotations_HBB, annotations_OBB, CDSI_construction_scripts) are under CC BY 4.0 license.

VME_CDSI_datasets.zip has four components:

1. annotations_OBB: It holds TXT files in YOLO format with Oriented Bounding Box (OBB) annotations. Each annotation file is named after the corresponding image name
2. annotations_HBB: This component contains HBB annotation files in JSON file formatted in MS-COCO format defined by four values in pixels (x_min, y_min, width, height) of training, validation, and test splits 
3. satellite_images: This folder consists of VME images of size 512x512 in PNG format
4. CDSI_construction_scripts: This directory comprises all instructions needed to build the CDSI dataset in detail: a) instructions for downloading each dataset from its repository, b) The conversion to MS-COCO format script for each dataset is under the dataset name folder, c) The combination instructions. The training, validation, and test splits are available under "CDSI_construction_scripts/data_utils" folder.

annotations_HBB, annotations_OBB, CDSI_construction_scripts, are available in our GitHub repository

Please cite our dataset & paper with the preferred format as shown in the "Citation" section

@article{al-emadi_vme_2025,
    title = {{VME: A Satellite Imagery Dataset and Benchmark for Detecting Vehicles in the Middle East and Beyond}},
    volume = {12},
    issn = {2052-4463},
    url = {https://doi.org/10.1038/s41597-025-04567-y},
    doi = {10.1038/s41597-025-04567-y},
    pages = {500},
    number = {1},
    journal = {Scientific Data},
    author = {Al-Emadi, Noora and Weber, Ingmar and Yang, Yin and Ofli, Ferda},
    date = {2025-03-25},
    publisher={Spring Nature},
    year={2025}
}