EQFace

EQFace: A Simple Explicit Quality Network for Face Recognition
The first face recognition network that generates explicit face quality online. It enables a lot of applications where face quality is used.

Requirements

  • Pytorch 1.8.1

Training Data

  1. Download MS1Mv2
  2. Extract image files by rec2image.py
  3. Generate the training file list
cd dataset
python generate_file_list.py

Test

  1. Download pretrained model
  2. run test_quality.py
python test_quality.py --backbone backbone.pth --quality quality.path --file test_faces

Training

Training pipeline

  1. Step 1: set config.py, then run python train_feature.py
    ...
    BACKBONE_RESUME_ROOT = ''
    HEAD_RESUME_ROOT = ''
    TRAIN_FILES = './dataset/face_train_ms1mv2.txt'
    BACKBONE_LR = 0.05
    PRETRAINED_BACKBONE = ''
    PRETRAINED_QUALITY = ''
    ...
  1. Step 2: set config.py, then run python train_quality.py
    ...
    BACKBONE_RESUME_ROOT = './backbone_resume.pth'
    HEAD_RESUME_ROOT = './head_resume.pth'
    TRAIN_FILES = './dataset/face_train_ms1mv2.txt'
    BACKBONE_LR = 0.05
    PRETRAINED_BACKBONE = ''
    PRETRAINED_QUALITY = ''
    ...
  1. Step 3: set config.py, then run python train_feature.py
    ...
    BACKBONE_RESUME_ROOT = ''
    HEAD_RESUME_ROOT = ''
    TRAIN_FILES = './dataset/face_train_ms1mv2.txt'
    BACKBONE_LR = 0.05
    PRETRAINED_BACKBONE = ''
    PRETRAINED_QUALITY = ''

    PRETRAINED_BACKBONE = 'pretrained_backbone_resume.pth'
    PRETRAINED_QUALITY = 'pretrained_qulity_resume.pth'
    ...

Performance Benchmark

Face verification on still image and TF video datasets
1:1 verification on IJB-B and IJB-C datasets
1:N identification on IJB-B and IJB-C datasets

Citation

  • If you think this work is useful for you, please cite

    @inproceedings{EQFace,
title = {EQFace: A Simple Explicit Quality Network for Face Recognition},
author = {Liu, Rushuai and Tan, Weijun},
booktitle = {CVPRW},
year = {2021}
}

GitHub

https://github.com/deepcam-cn/FaceQuality