YOLO bounding-box annotations for objects and accessibility-related elements in outdoor urban scenes.
This repository contains 2,988 label files, a nine-class configuration, eight preview images, a license and citation metadata. The full train/validation/test images are excluded by .gitignore and are not included in this checkout. Obtain the matching images from the dataset author before training; a full-image download URL is not currently documented here.
The following counts are calculated from the non-empty annotation lines currently committed in each split:
| Split | Label files | Bounding boxes |
|---|---|---|
| Train | 2,122 | 6,269 |
| Validation | 449 | 1,306 |
| Test | 417 | 1,282 |
| Total | 2,988 | 8,857 |
Earlier documentation reported 9,089 annotations. The table above describes the files in this repository; reconciling the original export requires the source dataset. Image resolution and image-to-label completeness cannot be verified without the full images.
The order below matches data.yaml. Class IDs are zero-based.
| ID | Class | ID | Class | ID | Class |
|---|---|---|---|---|---|
| 0 | car | 3 | obstacle | 6 | stairs |
| 1 | chair | 4 | person | 7 | trash |
| 2 | hole | 5 | sidewalk | 8 | tree |
train/labels/ Training annotations
valid/labels/ Validation annotations
test/labels/ Test annotations
preview/ Eight example images
data.yaml Split paths and ordered class names
CITATION.cff Machine-readable citation
LICENSE CC BY 4.0 license
Each label line contains:
class_id x_center y_center width height
Coordinates are normalized relative to the corresponding image. Match image and label basenames when restoring the dataset. The current YAML uses relative paths such as ../train/images; configure the dataset root and split paths for your training environment before use.
See preview/ for the remaining examples.
Computer vision research, object detection experiments and outdoor scene understanding. Accessibility-related categories describe scene objects; they are not a validated navigation or safety assessment.
@dataset{boarat2026oaodd,
author = {Dana Bo'arat},
title = {Outdoor Accessibility Object Detection Dataset (OAODD)},
year = {2026},
version = {1.0}
}See CITATION.cff and the Creative Commons Attribution 4.0 license.
Contact: Dana Bo'arat — danaboayrat@gmail.com