cazzz307/yolov8-crack-detection

YOLOv8 Crack Detection Model 🏗️

A YOLOv8 nano model fine-tuned for detecting cracks in infrastructure (walls, roads, concrete structures).

📸 Example Prediction

The model detects cracks in infrastructure with bounding boxes and confidence scores.

Model Description

This model detects cracks in images of buildings, roads, bridges, and other infrastructure.

Use Cases:

• 🏗️ Automated infrastructure inspection
• 🛣️ Road maintenance monitoring
• 🏢 Building safety assessment
• 🌉 Bridge and tunnel inspection
• 🚁 Drone-based crack detection

Model Details

• Model Type: YOLOv8 Nano (object detection)
• Task: Infrastructure crack detection
• Framework: Ultralytics YOLOv8
• Input Size: 640x640 pixels
• Classes: 1 (crack)

Training Data

• Training Images: 3,717
• Validation Images: 200
• Test Images: 112

Usage

Quick Start

from ultralytics import YOLO
from huggingface_hub import hf_hub_download

# Download model from Hugging Face
model_path = hf_hub_download(
    repo_id="cazzz307/yolov8-crack-detection",
    filename="best.pt"
)

# Load and use model
model = YOLO(model_path)
results = model.predict("crack_image.jpg", conf=0.25)
results[0].show()

Batch Processing

# Process multiple images
results = model.predict(["img1.jpg", "img2.jpg", "img3.jpg"])

for result in results:
    boxes = result.boxes
    for box in boxes:
        conf = box.conf[0]
        print(f"Crack detected! Confidence: {conf:.2f}")

Real-time Detection

# Webcam
model.predict(source=0, show=True)

# Video file
model.predict(source="video.mp4", save=True)

Model Files

• best.pt - Best model checkpoint (use this)
• last.pt - Last training checkpoint
• results.png - Training metrics
• confusion_matrix.png - Performance analysis

Command Line Usage

# Install
pip install ultralytics

# Download and use
yolo detect predict model=best.pt source=image.jpg

Limitations

• Optimized for concrete/asphalt cracks
• Performance varies with lighting
• May miss very fine hairline cracks

License

MIT License

---

Built with Ultralytics YOLOv8 🚀