README.md

metadata

license: other
license_name: pi-lab-license-1.0
license_link: LICENSE

This is the official datasets for the paper: FaceMe: Robust Blind Face Restoration With Personal Identification

Dataset Overview

This repository contains the following five datasets:

1. FFHQ
2. FFHQRef
3. LFW-Test-Ref
4. Wider-Test-Ref
5. WebPhoto-Test-Ref

Dataset Usage

• FFHQ and FFHQRef: Used for training.
• LFW-Test-Ref, Wider-Test-Ref, and WebPhoto-Test-Ref: Used for inference.

Important Notes

Due to the large size of the FFHQRef dataset, it has been split into multiple parts. Before using this dataset, you need to merge the split files into a single .tar file and then extract it.

Steps to Merge the Split Files

Run the following command to merge all parts into a single file:

ls FFHQRef_part_*.tar | sort -V | xargs cat > merged_FFHQRef.tar

After merging, extract the merged_FFHQRef.tar file to access the dataset:

tar -xvf merged_FFHQRef.tar