Awesome Face Recognition

[2020. 02. 02]

• Add Decoupled Network. (in model/dc_module.py)

Implementation

Datasets

- Training

• Faces-emore

- Evaluation

• LFW
• Agedb-30
• CFP-FP

Implemented Networks

• [CVPR 2018] Decoupled Networks
• [CVPR 2019] Arcface
• [CVPR 2019] Regularface

How to train ?

1. Set-up virtual environment.

   conda create -n [your environment] python=3.6
   source activate [your environment]
   conda install pytorch cuda90 torchvision -c pytorch
   pip install -r requirements.txt

   ​

1. Download train, test datasets
   ( https://github.com/TreB1eN/InsightFace_Pytorch —> 3.1 Data Preparation )

2. Training
   python train.py --train_root [your data root] --epochs [epochs] --batch_size [batch_size] --save_root [root for saving weights, log file]

Results

• Accuracy : LFW (VGG-19)
  

​

• Accuracy : AGE-DB30 (VGG-19)
  

References

1. https://github.com/TreB1eN/InsightFace_Pytorch
2. https://github.com/ronghuaiyang/arcface-pytorch
3. https://github.com/wy1iu/DCNets