Photo Restoration Python script

One of those 'scratching my own itch' type of Github repo. A comprehensive solution for restoring and enhancing old photographs using open-source AI models on rented GPUs. This suite combines colorization and face restoration techniques to breathe new life into your old family photos.

Just did a batch of c. 350 family photos and that costed me less than 1 dollar on a expensive datacenter Nvidia L40S GPU (USD 3,51 per hour) via Replicate.

Features

• Photo Colorization:
  • Colorize black & white photos
  • Convert red-tinted photos (degraded film) to black & white, then colorize them
• Face Restoration:
  • Enhance and restore face details using GFPGAN model
  • Fix blurry, low-quality, or damaged facial features
• Batch Processing:
  • Process multiple photos at once
  • Progress tracking with status updates
• Docker Support:
  • Easy deployment with Docker, so you do not have to mess around dependecies and that kind of stuff
  • Consistent environment across platforms

Output

After processing completes, you'll find:

• Colorized photos in the colorized/ directory
• Restored photos in the gfpgan/ directory
• Original photos remain unchanged in their respective directories

Models Used

• Colorization: DDColor model via Replicate API: https://replicate.com/piddnad/ddcolor
• Face Restoration: GFPGAN v1.4 via Replicate API: https://replicate.com/tencentarc/gfpgan

Running with Docker

To run the application using Docker:

1. Rename .env-template to .env using the following command, then add your Replicate API token to the .env file:

   mv .env-template .env

2. Place your photos in the appropriate folders:

   • Black & white photos go in photos/bn/
   • Red-tinted photos go in photos/red/
   • Any other photos go in photos/

3. Build and start the container:

   docker compose up --build

   • Progress will be displayed in the console
   • Results will appear in the output directories
   • The process will take a few minutes to complete and the container will stop automatically

4. Enhanced results are saved in the gfpgan/ directory

Troubleshooting

1. API Token Issues:
   • Ensure your Replicate API token is valid
   • Check if the token is properly set in the .env file

How It Works

The restoration process happens in two stages:

1. Colorization Stage:

   • Black & white photos (in photos/bn/) are colorized
   • Red-tinted photos (or any other color resulting from chemical degradation of the film) (in photos/red/) are converted to B&W, then colorized
   • Results are saved in both colorized/ and photos/ directories

2. Face Restoration Stage:

   • All photos (including newly colorized ones) are processed with GFPGAN model