metadata

tags:
  - super-resolution
  - image-enhancement
  - pytorch
  - deep-learning
datasets:
  - DIV2K
  - Set5
license: mit
model-index:
  - name: EDSR
    results:
      - task:
          name: Image Super-Resolution
          type: image-super-resolution
        dataset:
          name: DIV2K
          type: DIV2K
        metrics:
          - name: PSNR
            type: psnr
            value: 30.5
          - name: SSIM
            type: ssim
            value: 0.85

Super Resolution Model Comparison

🚀 Super Resolution Model Comparison is a web application built using Streamlit that allows users to upload a low-resolution image and compare its enhancement using various super-resolution models (SRCNN, VDSR, and EDSR). The application provides image quality metrics (PSNR and SSIM) and processing time for each model to assess their performance.

Features

Image Upload: Upload low-resolution images in PNG, JPG, or JPEG formats.
Model Comparison: Enhance the uploaded image using three popular super-resolution models:
- SRCNN: Super-Resolution Convolutional Neural Network.
- VDSR: Very Deep Super-Resolution Network.
- EDSR: Enhanced Deep Super-Resolution Network.
Visual Comparison: View the enhanced output for each model side-by-side.
Performance Metrics:
- Processing Time: Time taken by each model to process the image.
- Image Quality:
  - PSNR (Peak Signal-to-Noise Ratio): Measures the quality of the enhanced image compared to the original.
  - SSIM (Structural Similarity Index): Assesses structural similarity between the images.
Download Enhanced Images: Download the enhanced images generated by each model.

Installation

Prerequisites

Python 3.8 or above
Pip package manager

Steps

Clone the Repository

git clone <repository-url>
cd <repository-folder>

Install Dependencies
```
pip install -r requirements.txt
```
Download Pre-trained Weights
- Place the pre-trained model weights in the checkpoints/ directory:
  - srcnn_best.pth for SRCNN
  - vdsr_best.pth for VDSR
  - edsr_best.pth for EDSR
Run the Application
```
streamlit run app.py
```

Usage

Open the application in your browser (default URL: http://localhost:8501).
Upload a low-resolution image using the file uploader.
View the enhanced images generated by each model.
Compare the performance and quality metrics for each model.
Download the enhanced images for further use.

File Structure

.
├── app.py                # Main Streamlit application script
├── models/               # Directory for model definitions
│   ├── srcnn.py          # SRCNN model
│   ├── vdsr.py           # VDSR model
│   └── edsr.py           # EDSR model
├── checkpoints/          # Directory for storing pre-trained weights
│   ├── srcnn_best.pth
│   ├── vdsr_best.pth
│   └── edsr_best.pth
├── requirements.txt      # Python dependencies
└── README.md             # Documentation

Dependencies

Install the required Python libraries:

torch: PyTorch for deep learning.
torchvision: Image transformations and utilities.
streamlit: Interactive web interface.
pillow: Image processing library.
numpy: Numerical operations.
scikit-image: Image quality metrics (PSNR, SSIM).

Models

1. SRCNN

A shallow neural network for super-resolution.
Faster but may have lower performance on complex images.

2. VDSR

A deeper network providing improved results at the cost of longer processing time.

3. EDSR

Enhanced deep network specifically designed for high-quality super-resolution.

Metrics

PSNR: Measures the similarity of the enhanced image to the original in terms of pixel-level accuracy. Higher values indicate better quality.
SSIM: Evaluates the structural similarity of the enhanced image to the original. Higher values indicate better preservation of structural features.

Customization

Adding New Models

Add the new model definition to the models/ directory.
Update the load_model() function in app.py to include the new model.

Updating Pre-trained Weights

Place updated weights in the checkpoints/ directory with the naming convention <model_name>_best.pth.

Future Enhancements

Support for real-time video super-resolution.
Integration with additional state-of-the-art super-resolution models.
GPU acceleration for faster processing.

License

This project is open-source and available under the MIT License.

Contact

For queries or contributions, reach out via GitHub or email.

Happy enhancing! 🚀