---
title: MFIR - Multi-Frame Image Restoration
emoji: 🖼️
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: 4.44.0
app_file: app.py
pinned: false
license: mit
---

# MFIR - Multi-Frame Image Restoration

A deep learning model that combines multiple degraded frames into a single high-quality restored image.

## How It Works

1. **Upload** 2-16 frames of the same scene (can have different degradations)
2. **Select** processing size and reference frame
3. **Click** "Restore Image" to get the fused output

## Architecture

- **Shared Encoder**: Extracts features from each frame
- **Deformable Alignment**: Aligns all frames to reference using learned offsets
- **Temporal Attention Fusion**: Fuses aligned features using multi-head attention
- **Decoder**: Reconstructs the final high-quality image

## Links

- [GitHub Repository](https://github.com/marduk-ra/MFIR)
- [Model Weights](https://huggingface.co/marduk-ra/MFIR)

## Author

**Veli Karaarslan** - 2026

## License

MIT License