This repository contains information about ParkSeg12k, an open-source dataset developed for semantic segmentation of parking lots using satellite imagery. The dataset was created to address the scarcity of publicly available data for analyzing off-street parking lots in the United States.
- 12,617 Image-Mask Pairs: Includes 512x512-pixel imagery, each annotated with parking lot boundaries.
- Wide Geographic Coverage: Covers ∼35,000 parking lots across 45 U.S. cities, totaling 297.7 km² of area and 62.5 km² of labeled parking lots.
- Two Versions:
- 3-Channel RGB Dataset: Standard satellite imagery.
- Near-Infrared (NIR) Channel Dataset: Includes an additional fourth channel (NIR) corresponding to each 3-channel RGB image to enhance segmentation accuracy.
- Data Split: The dataset is divided into a 90%-10% train-test split.
The dataset is hosted on Hugging Face and can be downloaded here: UTEL-UIUC/parkseg12k
The annotation for the mask images in this dataset was created by correcting and processing the Parking Reform Network (PRN) data and OpenStreetMap (OSM) data to make it suitable for our task. These original datasets were adjusted to meet the specific requirements of our research.
If you use this dataset in your research, please cite the corresponding paper:
@article{qiam2024,
title={A Pipeline and NIR-Enhanced Dataset for Parking Lot Segmentation},
author={Shirin Qiam and Saipraneeth Devunuri and Lewis J. Lehe},
journal={arXiv preprint arXiv:2412.13179},
year={2024},
url={https://arxiv.org/pdf/2412.13179}
}Note: This citation will be updated with the final publication details once the paper is officially published.