🔍 Concrete Crack Segmentation with UNet

Upload an image of a concrete surface to detect and segment cracks using a trained UNet model.

Features:

  • Advanced UNet architecture with batch normalization and dropout
  • Optimized for highly imbalanced crack detection
  • Interactive threshold adjustment
  • Colored overlay visualization

How to use:

  1. Upload a concrete surface image
  2. Click "Detect Cracks" to run the segmentation
  3. View the results: white areas in the mask indicate detected cracks
  4. Adjust the threshold in Advanced mode for fine-tuning sensitivity

Model Information:

  • Architecture: Improved UNet with BatchNorm and Dropout
  • Input Size: Images are resized to 128x128 for processing
  • Output: Binary segmentation mask highlighting crack regions
  • Training: Optimized for imbalanced crack detection using advanced loss functions

Tips for better results:

  • Use high-contrast images where cracks are visible
  • Ensure good lighting conditions
  • Try adjusting the threshold if results seem too sensitive or not sensitive enough