Case study 04
VISION AI
Restore low-quality images and videos with super-resolution and face enhancement.
Overview
VISION AI is an image and video restoration tool built as a hackathon project. It upscales low-resolution images and repairs blurred or damaged faces in video, frame by frame.
The goal was to make state-of-the-art restoration models usable from a simple web interface, without any setup on the user's side.
What it does
Image super-resolution
ESRGAN upscales low-resolution images while recovering fine texture and edges.
Video enhancement
Videos are split into frames with OpenCV, enhanced, and stitched back together.
Face detection
RetinaFace locates every face in a frame so restoration is applied where it matters.
Face restoration
GFPGAN and CodeFormer rebuild facial detail in blurred or compressed footage.
Format conversion
Convert results between common image formats straight from the app.
Domain upscaling
Satellite and medical image upscaling are planned as dedicated modes.
How it works
- 01Upload an image or video from the React frontend.
- 02The Flask API routes it to the right pipeline.
- 03Images go through ESRGAN; videos are split into frames with OpenCV.
- 04Faces are detected with RetinaFace and restored with GFPGAN or CodeFormer.
- 05The enhanced result is returned for preview and download.
Stack
- Python
- PyTorch
- ESRGAN
- GFPGAN
- CodeFormer
- RetinaFace
- OpenCV
- Flask
- React
- Tailwind CSS