Bangladesh Road Damage Detection Dataset
Description
This dataset contains annotated road images collected from different roads in Bangladesh for road damage and pothole detection research. The dataset was prepared for object detection tasks using YOLO-compatible annotation format. The dataset includes both damaged road images and normal road images to improve model generalization and reduce false positive predictions. Road damages such as potholes, broken road surfaces, and visible cracks were annotated under a single class named "pothole". The dataset is divided into train, validation, and test sets and exported in YOLOv8 format. It can be used for computer vision research, intelligent transportation systems, automated road monitoring, and deep learning-based road damage detection applications.
Files
Steps to reproduce
The data collection and preparation process followed these steps: Data Acquisition: Images were captured using three different mobile device cameras to ensure diversity in image quality, resolution, and sensor characteristics. Environmental Conditions: Photos were taken under various daylight conditions across different road types in Bangladesh, including highways, urban roads, and rural paths. Class Labeling: Road damages such as potholes, cracks, and broken surfaces were identified and manually annotated under a single unified class named "pothole" to simplify detection for specific use cases. Formatting: The annotated images were then processed and exported into the YOLO (You Only Look Once) format, making the dataset ready for training deep learning models.
Institutions
- Khwaja Yunus Ali UniversityRajshahi Division, Chauhāli