DAEFR: Degradation-Aware Face Restoration

This model checkpoint is for DAEFR (Degradation-Aware Face Restoration) - a face restoration model that handles various degradation levels.

Model Description

DAEFR is a degradation-aware face restoration framework that:

• Uses dual codebooks for high-quality and low-quality face restoration
• Employs an association stage to bridge HQ and LQ domains
• Achieves state-of-the-art results on blind face restoration benchmarks

Usage

from huggingface_hub import hf_hub_download
import torch

# Download checkpoint
checkpoint_path = hf_hub_download(
    repo_id="{repo_id}",
    filename="{checkpoint_name}"
)

# Load model
model = torch.load(checkpoint_path, map_location='cpu')

Training Details

• Epochs: 100
• Dataset: FFHQ 512x512
• Degradation: Synthetic blind degradation (blur, noise, JPEG, downsampling)

Citation

@article{DAEFR,
  title={Degradation-Aware Face Restoration},
  author={},
  journal={},
  year={2024}
}