# Drone-based AI Bridge Inspection Dataset (Kansas NCRPC SMART Stage 1)

## Overview
This dataset contains results from the Kansas NCRPC SMART Stage 1 project, where drone imagery and videos of county bridges were processed with an AI model to detect eight categories of concrete defects. The public dataset includes annotated image and video results and TXT defect count summaries.

## Defect Categories
The class labels below are retained exactly as they appear in the model outputs. The descriptions follow the terminology used in the associated paper.

| Class label | Defect name | Description |
| --- | --- | --- |
| Crack | Crack | A linear fracture in concrete that may be structural (load-induced) or non-structural (associated with temperature or shrinkage), distinguished from the interconnected pattern labeled ACrack. |
| ACrack | Alligator Crack | An interconnected, map-like network of cracks associated with shrinkage or temperature effects and described as typically non-structural in the paper. |
| Efflorescence | Efflorescence | White crystalline deposits on the concrete surface caused by leaching of calcium compounds through moisture movement. |
| WConccor | Water-Induced Concrete Washouts, Erosion and Corrosion | Deterioration of concrete due to prolonged exposure to water, often leading to surface degradation. |
| Spalling | Spalling | Localized breaking, flaking, or chipping of concrete, often exposing aggregates or reinforcement. |
| Wetspot | Wet spot | Visible darkened or damp areas on the concrete surface due to moisture accumulation. |
| Rust | Rust | Reddish-brown discoloration on concrete or steel surfaces caused by corrosion of steel. |
| ExposedRebars | Exposed Reinforcement Bars | Visible steel reinforcement due to loss of concrete cover. |

## Contents
The dataset is distributed across four ZIP archives:

- `BridgeDefects_Inference_Images_2025.zip`: Annotated image results, matching TXT defect count summaries, this README, and the license file.
- `BridgeDefects_Inference_Videos1_2025.zip`: Annotated video results, part 1.
- `BridgeDefects_Inference_Videos2_2025.zip`: Annotated video results, part 2.
- `BridgeDefects_Inference_Videos3_2025.zip`: Annotated video results, part 3.

### Image Archive Structure
Within the image archive:

- `Images/`: Contains annotated JPEG images with overlaid detection polygons and class labels, together with matching `_counts.txt` files containing per-image defect count summaries.
- `README.md`: Dataset description, defect category definitions, and usage instructions.
- `LICENSE.txt`: License information.

Each image and its corresponding TXT file share a filename stem. For example, `Images/DSC02807.jpg` is paired with `Images/DSC02807_counts.txt`.

The TXT files are stored alongside the images in `Images/`; there is no separate `text_counts/` folder. JSON outputs are not included in this public dataset, and there is no `json/` folder. The video results are distributed in the separate video archives listed above, rather than in a `videos/` folder within the image archive.

## Usage
- Annotated images: View the AI-detected defect locations and class labels.
- Annotated videos: View AI detections across video frames.
- TXT files: Read per-image defect count summaries for comparison alongside the matching annotated images.

## Citation
If you use this dataset, please cite:
**“Drone-Based AI for Automated Bridge Defect Detection and Condition Assessment” (Natnael Taye et al., 2025).**

## License
This dataset is released under the **CC BY 4.0 license**. See LICENSE.txt for details.