🏗️ Concrete Crack Segmentation

AI-Powered Crack Detection for Infrastructure Inspection

Upload a concrete image to detect and segment cracks automatically!

  • Model: Improved UNet (4-level encoder-decoder)
  • Accuracy: 99.77% Dice Score
  • Speed: Real-time inference

📤 Input

0.1 0.9

📊 Results

ℹ️ How to Use

  1. Upload an image of concrete (JPG, PNG, etc.)
  2. Adjust the detection threshold if needed (0.1-0.9)
  3. Click "Detect Cracks" to process
  4. View results in different formats:
    • Original: Input image
    • Mask: Binary segmentation (white=cracks, black=background)
    • Visualization: Overlay with cracks highlighted in red

📋 Image Requirements

  • Format: JPG, PNG, BMP, WebP
  • Size: Recommended 256×256 or larger
  • Content: Concrete surfaces (walls, pavements, structures)
  • Lighting: Well-lit images for best results

⚙️ Technical Details

  • Model: UNet with 7.8M parameters
  • Input: RGB images (automatically resized to 256×256)
  • Output: Probability maps + binary masks
  • Processing Time: ~50-100ms per image
  • Device: GPU (CUDA) if available, otherwise CPU

🎯 Results Interpretation

Crack % Status Action
< 0.5% 🟢 Good No action needed
0.5-2% 🟡 Fair Monitor regularly
2-5% 🟠 Caution Schedule inspection
> 5% 🔴 Warning Urgent inspection required

Model Status: Production Ready ✓
Created by: samir-mohamed
License: MIT
Repository: https://github.com/samir-m0hamed/Concrete-Crack-Segmentation-via-UNet