Crack Segmentation Multi-Model Comparison

This Streamlit application allows users to compare multiple deep learning models for crack segmentation in images. Users can upload single or multiple images, or even a zip file containing images, and see the segmentation results from different models side by side.

Features

• Support for single image, multiple image, and zip file uploads
• Comparison of five different segmentation models:
  1. CrackFusionNet
  2. UnetSEResnet50: U-Net with SE-ResNet50 encoder
  3. UnetPlusPlusResNet18: U-Net++ with ResNet18 encoder
  4. DeepLabV3+: DeepLabV3+ with ResNet18 encoder
  5. FPN: Feature Pyramid Network with ResNet18 encoder
• Display of original images in a carousel
• Visualization of segmentation results for each model
• Sorting of results based on total predicted crack pixels
• Download options for segmentation results

Dataset