MFIR / README.md
marduk-ra's picture
Upload 2 files
5e01da9 verified

A newer version of the Gradio SDK is available: 6.13.0

Upgrade
metadata
title: MFIR - Multi-Frame Image Restoration
emoji: 🖼️
colorFrom: blue
colorTo: purple
sdk: gradio
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

Author

Veli Karaarslan - 2026

License

MIT License