README.md

metadata

title: Tunnel Crack Detection
emoji: 🔍
colorFrom: blue
colorTo: red
sdk: streamlit
sdk_version: 1.28.0
app_file: app.py
pinned: false
license: agpl-3.0

🔍 Tunnel Crack Detection

A state-of-the-art web application for detecting cracks and defects in tunnel infrastructure using YOLOv12-DINO deep learning model. Deployed on Hugging Face Spaces for easy access and sharing.

Features

• 🔍 Real-time crack detection using YOLOv12-DINO
• 🖼️ Interactive web interface built with Streamlit
• 📊 Detailed analysis with charts and detection tables
• 📁 Large file support (up to 5GB model weights)
• 📈 Detection history and results export
• 🎯 Configurable parameters (confidence, IoU thresholds)

🚀 Quick Start

This application is deployed on Hugging Face Spaces. Simply:

1. Upload your YOLOv12-DINO model weights (.pt file) using the sidebar
2. Upload a tunnel image for analysis
3. Click "Analyze for Cracks" to get instant results!

🛠️ Local Setup

Prerequisites

• Python 3.8 or higher
• YOLOv12-DINO trained model weights (.pt file)

Installation

1. Clone or navigate to the project directory:

   cd /Users/sompoteyouwai/env/deploy_tunnel_crack
   

2. Create a virtual environment (recommended):

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate
   

3. Install dependencies:

   pip install -r requirements.txt
   

Running the Application

1. Start the Streamlit app:

   streamlit run streamlit_app.py
   

2. Open your browser and navigate to http://localhost:8501

3. Load the model:

   • If the default model exists at /Users/sompoteyouwai/env/model_weight/segment_defect.pt, click "Load Default Model"
   • Alternatively, upload your own model weights (.pt file)

4. Upload an image and click "Detect Cracks" to analyze

Usage

Model Loading

• Default Model: The app automatically detects if the default model is available
• Custom Model: Upload your own YOLOv12-DINO trained weights
• Device Selection: Choose between CPU, CUDA (GPU), or MPS (Apple Silicon)

Detection Parameters

• Confidence Threshold: Minimum confidence score for detections (0.01-1.0)
• IoU Threshold: Intersection over Union threshold for Non-Maximum Suppression (0.01-1.0)
• Image Size: Input resolution for the model (320-1280 pixels)

Results

• Annotated Images: View detected cracks with bounding boxes
• Detection Summary: Count and classification of detected defects
• Interactive Charts: Pie charts and confidence distributions
• Detailed Table: Exportable CSV with detection coordinates and metadata
• History: Track multiple detection sessions

File Structure

deploy_tunnel_crack/
├── streamlit_app.py      # Main Streamlit application
├── inference.py          # YOLOv12-DINO inference engine
├── requirements.txt      # Python dependencies
├── README.md            # This file
└── run.sh               # Quick start script

Model Information

The application expects a YOLOv12-DINO model trained for crack/defect detection. The default model path is:

/Users/sompoteyouwai/env/model_weight/segment_defect.pt

Supported Model Features:

• Object detection and segmentation
• Multiple defect classes
• Configurable inference parameters
• GPU acceleration support

Troubleshooting

Common Issues:

1. Model not found: Ensure the model file exists at the specified path or upload a custom model
2. CUDA errors: Install appropriate PyTorch version for your GPU
3. Memory issues: Reduce image size or use CPU device for large images
4. Import errors: Ensure all dependencies are installed: pip install -r requirements.txt

Performance Tips:

• Use GPU (CUDA/MPS) for faster inference
• Optimize image size based on your hardware capabilities
• Adjust confidence threshold to filter false positives

Development

To modify or extend the application:

1. Streamlit App: Edit streamlit_app.py for UI changes
2. Inference Engine: Modify inference.py for model-related changes
3. Dependencies: Update requirements.txt for new packages

License

This project is for tunnel infrastructure monitoring and maintenance applications.