Attain: Inclusive annotated pavement distress types and severity dataset.

Area of Science:

• Civil Engineering
• Computer Science
• Artificial Intelligence

Background:

• Pavement distress detection is vital for effective pavement management and rehabilitation (M&R).
• Existing datasets may lack diversity in distress types, severity levels, or data collection methods.
• Automated systems require comprehensive datasets for robust model training.

Purpose of the Study:

• Introduce the Attain dataset for pavement distress classification and object detection.
• Facilitate the development of machine learning and deep learning models for pavement M&R.
• Provide a cost-effective, high-quality dataset using readily available smartphone cameras.

Main Methods:

• Collected 2293 images using smartphones mounted on vehicle windshields.
• Manually annotated images by pavement engineering experts, identifying 10 distress categories and 2 severity levels.
• Annotated 19,761 distress instances with type and severity.

Main Results:

• The Attain dataset features 10 pavement distress categories with low, medium, and high severity levels.
• Includes diverse conditions under various lighting and traffic scenarios.
• 19,761 distress instances are annotated, offering rich data for model training.

Conclusions:

• The Attain dataset supports the development of automated pavement distress detection systems.
• Enhances the efficiency and accuracy of pavement maintenance decision-making.
• Its diverse and detailed annotations make it a valuable resource for researchers and practitioners.