README.md

---
license: apache-2.0
tags:
- vision
- object-detection
- yolo
- crack-detection
- infrastructure
- civil-engineering
datasets:
- custom
metrics:
- mAP
- precision
- recall
---

# Crack Detection YOLO Model

This repository contains a YOLO-based object detection model specifically trained to identify and localize cracks in various infrastructure surfaces, including concrete walls, floors, and brick facades.

## Model Overview

- **Task**: Object Detection
- **Class**: `crack`
- **Architecture**: YOLO (Ultralytics)
- **Training Epochs**: 100
- **Input Resolution**: 640x640 (standard YOLO inference)

## Performance & Training Graphics

The following graphics demonstrate the model's training progression and final performance metrics.

### Training Results
![Training Results](9afc4826-bbbe-4107-a616-c63699267e78.png)
*The plots show consistent convergence in both training and validation losses (Box, Cls, DFL) over 100 epochs, with precision and recall stabilizing at high levels.*

### Precision-Confidence Curve
![Precision-Confidence Curve](f2f90a21-c218-4d7f-9ccf-39a3403386df.png)
*The model achieves a precision of 1.00 at a confidence threshold of 0.987, indicating very high reliability in detections at high confidence levels.*

## Inference Examples

The model has been validated on diverse surfaces showing robust detection capabilities.

### Concrete Surface Detection
![Concrete Crack Detection](5617d570-d2ad-485d-97dc-a9f2ed14acc3.jpeg)

### Building & Brick Wall Detection
![Brick Wall Crack Detection](99f3ecba-e888-4596-a131-a34f907c4d4a.jpeg)

## Usage

To use this model with the Ultralytics YOLOv8 library:

```python
from ultralytics import YOLO

# Load the model
model = YOLO('crack.pt')

# Perform inference on an image
results = model.predict('path/to/your/image.jpg', save=True, conf=0.5)

# View results
for result in results:
    result.show()
```

## Dataset Information

The model was trained on a specialized dataset focused on structural cracks. It includes variations in lighting, surface textures, and crack sizes to ensure better generalization in real-world infrastructure inspections.