/FaceQuality

An implementation of EQFace: A Simple Explicit Quality Network for Face Recognition (https://arxiv.org/abs/2105.00634, CVPRW 2021)

Primary LanguagePython

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.

Face Quality Result

License

The code of EQFace is released under the MIT License. There is no limitation for both academic and commercial usage.

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}
    }