BanglaCrack-10: A Real-World Bangladesh Building Crack Image Dataset with Multi-Class Bounding-Box Annotations
Description
BanglaCrack-10 is a real-world building crack and surface-condition image dataset collected in Bangladesh for computer vision and automated structural-surface inspection research. The released dataset contains 3,603 images collected from multiple locations in Bangladesh, including Adamdighi, Santahar, Chalk Sutarpur, Seujgari, and Shibganj in Bogura; and Tatibazar in Old Dhaka, Kuril, Khilkhet, and Bashundhara Residential Area in Dhaka. The dataset covers ten classes: Diagonal Cracks, Hairline Cracks, Horizontal Cracks, Map Pattern Cracks, No Crack, Shrinkage Cracks, Spalling Cracks, Stair Step Cracks, Structural Overload Cracks, and Vertical Cracks. Visible crack and surface-condition regions were manually annotated using axis-aligned bounding boxes in Roboflow. The audited YOLOv8 release contains 3,622 bounding-box annotations and is partitioned into 2,571 training images, 691 validation images, and 341 test images. All images in the released YOLOv8 version are standardized to 640 × 640 pixels. The dataset represents real building-surface conditions captured under natural environmental conditions, varying viewing angles, and different camera-to-surface distances. BanglaCrack-10 is intended to support research in object detection, crack classification, computer vision, structural health monitoring, automated building inspection, and transfer learning.
Files
Steps to reproduce
Download and extract the BanglaCrack-10 YOLOv8 dataset archive. The released dataset contains 3,603 images standardized to 640 × 640 pixels and corresponding YOLO-format annotation files. The dataset is organized into training, validation, and test partitions containing 2,571, 691, and 341 images, respectively. The data.yaml file defines the ten class names and class identifiers. Each YOLO annotation file contains the class ID followed by normalized bounding-box coordinates in the format: class_id x_center y_center width height. The dataset can be loaded using standard YOLOv8-compatible training or evaluation pipelines. When using the dataset in another environment, verify that the relative train, validation, and test paths in data.yaml correspond to the extracted dataset location. No additional preprocessing is required to access the released 640 × 640 images and annotations.
Institutions
- American International University-BangladeshDhaka Division, Dhaka