Skip to content

PyTorch Face Recognizer based on 'VGGFace2: A dataset for recognising faces across pose and age'

License

Notifications You must be signed in to change notification settings

cydonia999/VGGFace2-pytorch

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

4 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

PyTorch Face Recognizer based on 'VGGFace2: A dataset for recognising faces across pose and age'.

This repo implements training and testing models, and feature extractor based on models for VGGFace2 [1].

Pretrained models for PyTorch are converted from Caffe models authors of [1] provide.

Dataset

To download VGGFace2 dataset, see authors' site.

Preprocessing images

Faces should be detected and cropped from images before face images are fed to this face recognizer(demo.py).

There are several face detection programs based on MTCNN [3].

Pretrained models

The followings are PyTorch models converted from Caffe models authors of [1] provide.

arch_type download link
resnet50_ft link
senet50_ft link
resnet50_scratch link
senet50_scratch link

Extracting features

Usage:

python demo.py extract <options>

Options

  • --arch_type network architecture type (default: resnet50_ft):
    • resnet50_ft ResNet-50 which are first pre-trained on MS1M, and then fine-tuned on VGGFace2
    • senet50_ft SE-ResNet-50 trained like resnet50_ft
    • resnet50_scratch ResNet-50 trained from scratch on VGGFace2
    • senet50_scratch SE-ResNet-50 trained like resnet50_scratch
  • --weight_file weight file converted from Caffe model(see here)
  • --resume checkpoint file used in feature extraction (default: None). If set, --weight_file is ignored.
  • --dataset_dir dataset directory
  • --feature_dir directory where extracted features are saved
  • --test_img_list_file image file for which features are extracted
  • --log_file log file
  • --meta_file Meta information file for VGGFace2, identity_meta.csv in Meta.tar.gz
  • --batch_size batch size (default: 32)
  • --gpu GPU devide id (default: 0)
  • --workers number of data loading workers (default: 4)
  • --horizontal_flip horizontally flip images specified in --test_img_list_file

Testing

Usage:

python demo.py test <options>

Options

  • --arch_type network architecture type (default: resnet50_ft):
    • resnet50_ft ResNet-50 which are first pre-trained on MS1M, and then fine-tuned on VGGFace2
    • senet50_ft SE-ResNet-50 trained like resnet50_ft
    • resnet50_scratch ResNet-50 trained from scratch on VGGFace2
    • senet50_scratch SE-ResNet-50 trained like resnet50_scratch
  • --weight_file weight file converted from Caffe model(see here)
  • --resume checkpoint file used in test (default: None). If set, --weight_file is ignored.
  • --dataset_dir dataset directory
  • --test_img_list_file text file containing image files used for validation, test or feature extraction
  • --log_file log file
  • --meta_file Meta information file for VGGFace2, identity_meta.csv in Meta.tar.gz
  • --batch_size batch size (default: 32)
  • --gpu GPU devide id (default: 0)
  • --workers number of data loading workers (default: 4)

Training

Usage:

python demo.py train <options>

Options

  • --arch_type network architecture type (default: resnet50_ft):
    • resnet50_ft ResNet-50 which are first pre-trained on MS1M, and then fine-tuned on VGGFace2
    • senet50_ft SE-ResNet-50 trained like resnet50_ft
    • resnet50_scratch ResNet-50 trained from scratch on VGGFace2
    • senet50_scratch SE-ResNet-50 trained like resnet50_scratch
  • --weight_file weight file converted from Caffe model(see here), and used for fine-tuning
  • --resume checkpoint file used to resume training (default: None). If set, --weight_file is ignored.
  • --dataset_dir dataset directory
  • --train_img_list_file text file containing image files used for training
  • --test_img_list_file text file containing image files used for validation, test or feature extraction
  • --log_file log file
  • --meta_file Meta information file for VGGFace2, identity_meta.csv in Meta.tar.gz
  • --checkpoint_dir checkpoint output directory
  • --config number of settings and hyperparameters used in training
  • --batch_size batch size (default: 32)
  • --gpu GPU devide id (default: 0)
  • --workers number of data loading workers (default: 4)

Note

VGG-Face dataset, described in [2], is not planned to be supported in this repo. If you are interested in models for VGG-Face, see keras-vggface.

References

  1. ZQ. Cao, L. Shen, W. Xie, O. M. Parkhi, A. Zisserman, VGGFace2: A dataset for recognising faces across pose and age, 2018.
    site, arXiv

  2. Parkhi, O. M. and Vedaldi, A. and Zisserman, A., Deep Face Recognition, British Machine Vision Conference, 2015. site

  3. K. Zhang and Z. Zhang and Z. Li and Y. Qiao, Joint Face Detection and Alignment Using Multitask Cascaded Convolutional Networks, IEEE Signal Processing Letters, 2016. arXiv

About

PyTorch Face Recognizer based on 'VGGFace2: A dataset for recognising faces across pose and age'

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages