/focusface

Primary LanguagePython

Overall Test Dataset Link : https://drive.google.com/drive/folders/1GaGQOeaM-UlPuO0eLnFDxrxEJCG7J-V9?usp=sharing

DEMO Video VideoWriter: https://drive.google.com/file/d/150ASIsI86AY0qonqZwaCsHAXlMc3fEGG/preview

focusface

■ RTCDS_Project
├ ■ data
┃ ├ ■ weights
┃ ┃ ├ □ mobilenet0.25_Final.pth
┃ ┃ ├ □ mobilenetV1X0.25_pretrain.tar
┃ ┃ └ □ Resnet50_Final.pth
┃ ┃ 
┃ └ ■ target
┃   ├ ★ suspect_db_example.csv
┃   ├ □ [target_faces_dir]
┃   ┃ ├ □ [face_name_dir_001]
┃   ┃ ┃ ├ □ [face_name].jpg	
┃   ┃ ┃ ├ ☆ [face_name].[model].ebd
┃   ┃ ┃ └ ☆ [face_name].[model].checksum
┃   ┃ ┃
┃   ┃ ├ □ [face_name_dir_002]
┃   ┃ ：
┃   ┃ ├ □ [face_name_dir_003]
┃   ┃ └ □ [face_name_dir_004]
┃   ┃
┃   └ □ [sample_faces_dir]
┃   
└ ■ focusface
  ├ general
  ┃ └ ★ config.py
  ├ cmdl_stream_server.py
  ├ pyqt_stream_server.py
  └ focusface.ui

  1. Construct the directory path of RTCDS_Project (dirs marked with char '■' are required to be initilized with user's needs).

  2. Download the required data (marked with char '□')

  3. Check and modify configurations (marked with ★). You can refer to focusface/identifier/models/embedding_loader.py to create a initial csv file.

  4. After installation and setting required for execution, run one of ****_stream_server.py.

Installation

https://github.com/biubug6/Pytorch_Retinaface  # for face detection
https://github.com/ageitgey/face_recognition  # for face identification
https://github.com/abewley/sort  # for face tracking

Requirements

Installation Manual

  1. Create conda environment and Set python version
$ conda create -n [env_name] python=3.8
  1. Install cmake and dlib
Windows : https://github.com/ageitgey/face_recognition/issues/175#issue-257710508
osX(macOS), Linux : https://gist.github.com/ageitgey/629d75c1baac34dfa5ca2a1928a7aeaf
  1. Activate conda environment and Install Pytorch
# https://pytorch.org/get-started/previous-versions/

(example of windows, torch 1.7.0, cudatoolkit 10.1)
$ conda install pytorch==1.7.0 torchvision==0.8.0 torchaudio==0.7.0 cudatoolkit=10.1 -c pytorch
  1. Install other python libraries
$ pip install -r requirements.txt

Loading Processes

  1. Load Detector by importing module detector
    • Set fixed input resolution and disable training mode (set grad false)
    • Loads the network (re50 or mnet) onto the device (cpu or cuda) according to the entered parameters.
  2. Load Identifier by importing module identifier
    • In embedding_loader.py, read the DB (*.csv) to load the target's face embedding.
    • In the process of reading the embedding, an integrity check is performed by automatically matching sha256 hash.
    • Regenerate face embeddings that are defective as a result of the integrity check.
    • Pass the DB load result to the FaceComparer object and complete the load.
  3. Load SORT and set cv2.VideoCapture.

Arguments

refer to /general/configs.py

# at general/configs.py
# default value

cfg_opt_dict = {
    'data': '../data',
    'vid-res': 'adaptive',
    'det-model': 're50',
    'det-weight': 'weights/Resnet50_Final.pth',
    'box-ratio': 1.30,
    'down': 2,
    'conf-thresh': 0.50,
    'suspect-db': 'target/suspect_db_example.csv',
    's-faces': 'target/faces-400',
    'n-faces': 20,
    'idt-model': 'small',
    'iou-thresh': 0.30,
    'insense': 15,
    'criteria': 4.50,
    'redis-port': 6379,
    'output': 'opencv'
}

Citation

@inproceedings{deng2020retinaface,
title={Retinaface: Single-shot multi-level face localisation in the wild},
author={Deng, Jiankang and Guo, Jia and Ververas, Evangelos and Kotsia, Irene and Zafeiriou, Stefanos},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
pages={5203--5212},
year={2020}
}

@Article{dlib09,
author = {Davis E. King},
title = {Dlib-ml: A Machine Learning Toolkit},
journal = {Journal of Machine Learning Research},
year = {2009},
volume = {10},
pages = {1755-1758},
}