makeabilitylab/disabilityparking

AccessParkCV

<strong>AccessParkCV</strong> is a deep learning pipeline that detects and characterizes the width of disability parking spaces from orthorectified aerial imagery. We publish a dataset of 7,069 labeled parking spaces (and 4,693 labeled access aisles), which we used to train the models making AccessParkCV possible.

(This repo contains the data in a HuggingFace format. For raw COCO format, see link).

Dataset Description

This is an object detection dataset with 8 classes:

• —objects
• —access_aisle
• —curbside
• —dpnoaisle
• —dponeaisle
• —dptwoaisle
• —one_aisle
• —two_aisle

Dataset Structure

Data Fields

• —image: PIL Image object
• —width: Image width in pixels
• —height: Image height in pixels
• —objects: Dictionary containing:
• —bbox: List of bounding boxes in [xmin, ymin, xmax, ymax] format
• —category: List of category IDs
• —area: List of bounding box areas
• —iscrowd: List of crowd flags (boolean)
• —id: List of annotation IDs
• —segmentation: List of polygon segmentations (each as list of [x1,y1,x2,y2,...] coordinates)

Category IDs to Category

Data Sources

Class Composition

Data Splits

Class splits

Usage

from datasets import load_dataset

train_dataset = load_dataset("makeabilitylab/disabilityparking", split="train", streaming=True)

example = next(iter(train_dataset))

# Example of accessing an item
image = example["image"]
bboxes = example["objects"]["bbox"]
categories = example["objects"]["category"]
segmentations = example["objects"]["segmentation"]  # Polygon coordinates

Citation

@inproceedings{hwang_wherecanIpark,
  title={Where Can I Park? Understanding Human Perspectives and Scalably Detecting Disability Parking from Aerial Imagery},
  author={Hwang, Jared and Li, Chu and Kang, Hanbyul and Hosseini, Maryam and Froehlich, Jon E.},
  booktitle={The 27th International ACM SIGACCESS Conference on Computers and Accessibility},
  series={ASSETS '25},
  pages={20 pages},
  year={2025},
  month={October},
  address={Denver, CO, USA},
  publisher={ACM},
  location={New York, NY, USA},
  doi={10.1145/3663547.3746377},
  url={https://doi.org/10.1145/3663547.3746377}
}