Face recognition

Requirements

Python 3.5+

Commanline:

pip3 install -r requirements.txt

Usage:

Download

git clone https://github.com/vanlong96tg/Face_recognize_pytorch
mkdir face_recognize/weights
cd face_recognize/weights
wget https://www.dropbox.com/s/akktsgxp0n8cwn2/model_mobilefacenet.pth?dl=0 -O model_mobilefacenet.pth
wget https://www.dropbox.com/s/kzo52d9neybjxsb/model_ir_se50.pth?dl=0 -O model_ir_se50.pth
wget https://www.dropbox.com/s/rxavczg9dlxy3a8/model_ir50.pth?dl=0 -O model_ir50.pth

Python:

Run with default threshold=1.2:

python3 face_verify.py -csv {path_sample submit_csv} -path {path_folder_image} -image {path_image}

Use model ir_se50 (slower but more accurate):

python3 face_verify.py -csv {path_sample submit_csv}  -path {path_folder_image} -image {path_image}

Use model MobileNet change config.py:

change args net_size in get_config with net_size='mobi'

Args get_config(mode = 'app', net_size = 'large', net_mode = 'ir_se', use_mtcnn = 1, threshold = 1.25) in config.py:

mode: for demo
net_size: 'large' = model SE_IR50, 'mobi' = model MobileNet
net_mode: for net_size='large' value in ['ir_se', 'ir']
use_mtcnn: 1 using mtcnn default: recommend
threshold: distance > threshold => unknow 

Use model mtcnn for face detection:

python3 face_verify.py -csv {path_sample submit_csv} -path {path_folder_image} -image {path_image} -user_mtcnn 1

Run on video:

python3 infer_on_video.py 

Video and Face bank.Download video, Face bank and extract in dir.

Docker:

Install docker

curl -fsSL https://get.docker.com -o get-docker.sh
sudo sh get-docker.sh
sudo groupadd docker
sudo usermod -aG docker $USER

Install docker-compose

pip3 install docker-compose

Run

docker-compose up --build -d

Test API with docker

Requirements: url not authenticate

Test with Postman

Two image detection

URL: http://localhost:8084/face_recognition

{
	"image_url_origin":"https://www.dropbox.com/s/vm8fvi9xdmjrdmr/PQH_0000.png?dl=0",
	"image_url_detection":"https://www.dropbox.com/s/vm8fvi9xdmjrdmr/PQH_0000.png?dl=0"
}

Multiple image

Please zip image to file

URL: http://localhost:8084/face_recognition_two_image

{
	"image_url":"https://www.dropbox.com/s/vm8fvi9xdmjrdmr/PQH_0000.png?dl=0",
	"file_zip_url":"https://www.dropbox.com/s/bf705wgk2n9vog6/test.zip?dl=0"
}

Training:

• Performance

  LFW	CFP_FF	AgeDB	Vggface2_FP
  99.73	99.68	97.32	94.88

Acknowledgement

• This repo is inspired by InsightFace.MXNet, InsightFace.PyTorch, ArcFace.PyTorch, MTCNN.MXNet and PretrainedModels.PyTorch.
• Training Datasets Dataset-Zoo