offical pytorch implement of seesawfacenet

1. Intro

• This repo is a reimplementation of seesawfacenet(paper)
• For models, including the pytorch implementation of the backbone modules of Arcface/MobileFacenet/seesawfacenet(including seesaw_shareFaceNet,seesaw_shuffleFaceNet, DW_seesawFaceNetv1, DW_seesawFaceNetv2)
• We build build this repo based on the work of @TreB1eN(https://github.com/TreB1eN/InsightFace_Pytorch), and fixed few bugs before usage.
• Pretrained models are posted, include the MobileFacenet in the original paper

2. Pretrained Models & training logs & Performance

seesawfacenet @ googledrive seesawfacenet @ baidudisk extraction code:exiy

3. How to use

• clone

  git clone https://github.com/TropComplique/mtcnn-pytorch.git
  

3.1 Data Preparation

3.1.1 Prepare Facebank (For testing over camera or video)

Provide the face images your want to detect in the data/face_bank folder, and guarantee it have a structure like following:

data/facebank/
        ---> id1/
            ---> id1_1.jpg
        ---> id2/
            ---> id2_1.jpg
        ---> id3/
            ---> id3_1.jpg
           ---> id3_2.jpg

3.1.2 download the pretrained model to work_space/model

If more than 1 image appears in one folder, an average embedding will be calculated

3.2.3 Prepare Dataset (MS1MV2(face_emore, refined MS1M...whatever we call it) For training refer to the original paper)

download the MS1MV2 dataset:

• emore dataset @ BaiduDrive, emore dataset @ Dropbox
• More Dataset please refer to the original post

Note: If you use MS1MV2 dataset and the cropped VGG2 dataset, please cite the original papers.

• after unzip the files to 'data' path, run :

  python prepare_data.py
  

  after the execution, you should find following structure:

faces_emore/
            ---> agedb_30
            ---> calfw
            ---> cfp_ff
            --->  cfp_fp
            ---> cfp_fp
            ---> cplfw
            --->imgs
            ---> lfw
            ---> vgg2_fp

3.2 detect over camera:

• 1. download the desired weights to model folder:

• 2 to take a picture, run

  python take_pic.py -n name
  

  press q to take a picture, it will only capture 1 highest possibility face if more than 1 person appear in the camera

• 3 or you can put any preexisting photo into the facebank directory, the file structure is as following:

- facebank/
         name1/
             photo1.jpg
             photo2.jpg
             ...
         name2/
             photo1.jpg
             photo2.jpg
             ...
         .....
    if more than 1 image appears in the directory, average embedding will be calculated

• 4 to start

  python face_verify.py 
  

3.3 detect over video:

​```
python infer_on_video.py -f [video file name] -s [save file name]
​```

the video file should be inside the data/face_bank folder

previous work on mtcnn for android platform and face cropping

• mtcnn_android_native mtcnn_android_native
• Face-extractor-based-on-mtcnn Face-extractor-based-on-mtcnn

3.4 Training:

​```
python train.py -b [batch_size] -lr [learning rate] -e [epochs]

# python train.py -net mobilefacenet -b 256 -w 24
​```

4. References

• This repo is mainly based on TreB1eN/InsightFace_Pytorch and cvtower/SeesawNet_pytorch, and inspired by deepinsight/insightface as well.

PS

• PRs are welcome, especially for models for mobile platfroms
• Email : [email protected]

Citation

Please cite our papers in your publications if it helps your research:

@misc{zhang2019seesawnet, title={Seesaw-Net: Convolution Neural Network With Uneven Group Convolution}, author={Jintao Zhang}, year={2019}, eprint={1905.03672}, archivePrefix={arXiv}, primaryClass={cs.CV} }

@misc{zhang2019seesawfacenets, title={SeesawFaceNets: sparse and robust face verification model for mobile platform}, author={Jintao Zhang}, year={2019}, eprint={1908.09124}, archivePrefix={arXiv}, primaryClass={cs.CV} }