BLACrack Dataset

BLACrack is a high-resolution asphalt pavement crack segmentation dataset collected from real-world highway inspection scenarios. It is designed to support research on robust pavement crack segmentation under complex illumination conditions, severe background interference, and large-scale road surface imagery.

A typical example from the BLACrack dataset is shown below.

Dataset Overview

The BLACrack dataset was collected using a highway detection vehicle operating at approximately 80 km/h on asphalt pavement in Liaoning Province, China. The acquisition system uses a high-resolution line-scan camera with vertical orthographic imaging, which effectively reduces perspective distortion and preserves the geometric properties of pavement cracks, such as crack width and length.

Compared with many existing public crack datasets, BLACrack provides high-resolution top-down pavement images with realistic highway disturbances, including low illumination, bright exposure, traffic markings, contamination spots, tree shadows, and complex pavement textures. These characteristics make BLACrack a challenging benchmark for tiny-crack-in-large-background segmentation.

Dataset Composition

BLACrack consists of three subsets:

Subset	Number of Images	Resolution	Description
BrightCrack	300	3100 × 2088	Crack images captured under bright illumination conditions
LowLightCrack	265	3100 × 2088	Crack images captured under low-light conditions
AlligatorCrack	165	3100 × 2088	Images containing alligator crack patterns

In total, the dataset contains 730 high-resolution pavement images with corresponding pixel-wise binary crack masks.

Annotation

All images are manually annotated with pixel-level binary masks. Multiple annotators participated in the labeling process, and the most consistently annotated samples were retained as ground-truth masks.

The annotation format follows a standard binary segmentation setting:

• foreground: crack pixels
• background: non-crack pavement regions

Key Characteristics

BLACrack has the following characteristics:

1. Real-world highway acquisition
   Images are captured from real asphalt pavements using a vehicle-mounted highway inspection system.

2. High-speed collection scenario
   The data are acquired while the inspection vehicle operates at approximately 80 km/h, reflecting practical highway inspection conditions.

3. High-resolution orthographic imaging
   Each image has a resolution of 3100 × 2088. The top-down imaging geometry helps preserve crack shape, width, and length.

4. Challenging crack segmentation setting
   The crack regions occupy only a very small proportion of the full image, resulting in a severe foreground-background imbalance.

5. Complex visual interference
   The dataset contains challenging disturbances such as shadows, low illumination, exposure variations, stains, traffic markings, and complex pavement textures.

Suggested Tasks

BLACrack is mainly designed for:

• pavement crack segmentation
• fine-structure segmentation
• high-resolution road damage perception
• robust segmentation under illumination and background interference
• crack geometry analysis and quantification

📍 Data Availability: The dataset will be publicly released under the CC BY-NC-SA 4.0 license upon official acceptance of the associated paper.

If you are interested in early access for academic research or potential collaboration, please feel free to contact the authors.

---

license: cc-by-nc-sa-4.0