DAEFR Core

Blind face restoration using Degradation-Aware Encoder-Face Restoration (DAEFR).

Installation

pip install -r requirements.txt

Usage

Python API

from daefr_core import load_model, restore_image, restore_folder

# Load model
model = load_model('model.ckpt')

# Restore single image
restore_image(model, 'input.jpg', 'output.jpg')

# Restore folder
restore_folder(model, 'input_folder/', 'output_folder/')

Command Line

# Single image
python daefr_restore.py --model model.ckpt --input photo.jpg --output restored.jpg

# Folder
python daefr_restore.py --model model.ckpt --input photos/ --output restored/

# CPU inference
python daefr_restore.py --model model.ckpt --input photo.jpg --output restored.jpg --device cpu

Input/Output

• Input: Any size image (resized to 512x512 internally)
• Output: 512x512 restored face image

API Reference

load_model(checkpoint, device='cuda')

Load a DAEFR model from checkpoint.

model = load_model('model.ckpt', device='cpu')

restore_image(model, input_path, output_path)

Restore a single image.

restore_image(model, 'input.jpg', 'output.jpg')

restore_folder(model, input_folder, output_folder)

Restore all images in a folder.

restore_folder(model, 'inputs/', 'outputs/')

License

Same as original DAEFR project.