Semantic Segmentation IoU Evaluation Dashboard

A professional web-based dashboard for evaluating semantic segmentation model performance using Intersection over Union (IoU) scores.

Features

✨ Professional Dashboard UI

• Modern dark theme with gradient styling
• Real-time IoU score visualization
• Class-wise performance metrics
• Responsive design for all devices

🚀 Easy to Use

• Single-click evaluation button
• Real-time progress feedback
• Beautiful animated results display

📊 Comprehensive Results

• Overall Mean IoU score
• Individual class-wise IoU scores
• Performance status indicators
• Visual progress bars for each class

Setup Instructions

1. Install Dependencies

pip install -r requirements.txt

2. Run the Application

python app.py

The application will start on http://localhost:5000

3. Access the Dashboard

Open your web browser and navigate to:

http://localhost:5000

Usage

1. Click the "Run Evaluation" button to start the model evaluation

2. The system will:

   • Load the trained model
   • Process the validation dataset
   • Calculate IoU scores for each class
   • Display results in real-time

3. View the results:

   • Overall Mean IoU at the top (large number)
   • Class-wise scores in the grid below
   • Each card shows the class name, percentage, and exact IoU value

Model Information

• Architecture: U-Net with ResNet18 backbone
• Classes: 10 semantic segments
  1. Trees
  2. Lush Bushes
  3. Dry Grass
  4. Dry Bushes
  5. Ground Clutter
  6. Flowers
  7. Logs
  8. Rocks
  9. Landscape
  10. Sky

Project Structure

├── app.py                      # Flask backend application
├── model.py                    # U-Net model architecture
├── dataset.py                  # Dataset loader
├── test.py                     # Original test script
├── train.py                    # Training script
├── duality_model_final.pth     # Trained model weights
├── requirements.txt            # Python dependencies
├── index.html                  # Dashboard HTML
├── styles.css                  # Dashboard styling
└── script.js                   # Dashboard interactions

Performance Metrics

The dashboard displays performance with visual indicators:

• 🟢 Excellent (≥ 80%): Very strong performance
• 🔵 Very Good (≥ 70%): Strong performance
• 🟢 Good (≥ 60%): Acceptable performance
• 🟡 Moderate (≥ 50%): Fair performance
• 🔴 Needs Improvement (< 50%): Poor performance

Troubleshooting

Issue: "Model not found" error

• Ensure duality_model_final.pth exists in the project root directory

Issue: "CUDA out of memory"

• The app will automatically fall back to CPU. This is slower but will work.

Issue: "No validation data found"

• Verify that the Data/val/Color_Images and Data/val/Segmentation directories exist and contain images

Technical Details

• Backend: Flask (Python web framework)
• Frontend: HTML5, CSS3, Vanilla JavaScript
• Deep Learning: PyTorch
• No external UI libraries: Pure CSS and vanilla JS for better performance

Author Notes

This dashboard provides an intuitive interface to visualize model performance. The IoU (Intersection over Union) metric is calculated as:

IoU = (Intersection Area) / (Union Area)

For each class, the dashboard averages the IoU scores across all validation images.

---

Happy Evaluating! 🚀