FaceRecognition-MTCNN-ArcFace

FaceRecognition with MTCNN using ArcFace

🚀 New Update (27-01-2023)

• Liveness Model:

  • Liveness detector capable of spotting fake faces and performing anti-face spoofing in face recognition systems
  • Our FaceRecognition system initially will check the faces are Fake or NOT
  • If its a Fake face it will give warnings
  • Otherwise it will go for Face-Recognition

Clone this Repository

git clone https://github.com/naseemap47/FaceRecognition-MTCNN-ArcFace.git
cd FaceRecognition-MTCNN-ArcFace

Install dependency

pip3 install -r requirements.txt

Custom Face Recognition

You can use:

• Command Line
  

• Streamlit Dashboard

Streamlit Dashboard

⚠️ New version NOT Available, Not updated Liveness Model

RUN Streamlit

streamlit run app.py

Command Line (Recommended)

1.Collect Data using Web-cam or RTSP

Args

-i, --source: RTSP link or webcam-id
-n, --name: name of the person
-o, --save: path to save dir
-c, --conf: min prediction conf (0<conf<1)
-x, --number: number of data wants to collect

Example:

python3 take_imgs.py --source 0 --name JoneSnow --save data --conf 0.8 --number 100

📖 Note:
Repeate this process for all people, that we need to detect on CCTV, Web-cam or in Video.
In side save Dir, contain folder with name of people. Inside that, it contain collected image data of respective people.
Structure of Save Dir:

├── data_dir
│   ├── person_1
│   │   ├── 1.jpg
│   │   ├── 2.jpg
│   │   ├── ...
│   ├── person_2
│   │   ├── 1.jpg
│   │   ├── 2.jpg
│   │   ├── ...
.   .
.   .

2.Normalize Collected Data

It will Normalize all data inside path to save Dir and save same as like Data Collected Dir

Args

-i, --dataset: path to dataset/dir
-o, --save: path to save dir

Example:

python3 norm_img.py --dataset data/ --save norm_data

Structure of Normalized Data Dir:

├── norm_dir
│   ├── person_1
│   │   ├── 1_norm.jpg
│   │   ├── 2_norm.jpg
│   │   ├── ...
│   ├── person_2
│   │   ├── 1_norm.jpg
│   │   ├── 2_norm.jpg
│   │   ├── ...
.   .
.   .

3.Train a Model using Normalized Data

Args

-i, --dataset: path to Norm/dir
-o, --save: path to save .h5 model, eg: dir/model.h5
-l, --le: path to label encoder
-b, --batch_size: batch Size for model training
-e, --epochs: Epochs for Model Training

Example:

python3 train.py --dataset norm_data/ --batch_size 16 --epochs 100

Inference

Args

-i, --source: path to Video or webcam or image
-m, --model: path to saved .h5 model, eg: dir/model.h5
-c, --conf: min prediction conf (0<conf<1)
-lm, --liveness_model: path to liveness.model
--le, --label_encoder: path to label encoder

On Image

Example:

python3 inference_img.py --source test/image.jpg --model models/model.h5 --conf 0.85 \
                     --liveness_model models/liveness.model --label_encoder models/le.pickle

To Exit Window - Press Q-Key

On Video or Webcam

Example:

# Video (mp4, avi ..)
python3 inference.py --source test/video.mp4 --model models/model.h5 --conf 0.85 \
                     --liveness_model models/liveness.model --label_encoder models/le.pickle

# Webcam
python3 inference.py --source 0 --model models/model.h5 --conf 0.85 \
                     --liveness_model models/liveness.model --label_encoder models/le.pickle

To Exit Window - Press Q-Key

🚀 Liveness Model

Liveness detector capable of spotting fake faces and performing anti-face spoofing in face recognition systems
If you wants to create a custom Liveness model, Follow the instruction below 👇:

Data Collection

Collect Positive and Negative data using data.py

Args

-i, --source: source - Video path or camera-id
-n, --name: poitive or negative

Example:

cd Liveness
python3 data.py --source 0 --name positive  # for positive
python3 data.py --source 0 --name negative  # for negative

Train Liveness Model

Train Liveness model using collected positive and negative data

Args

-d, --dataset: path to input dataset
-p, --plot: path to output loss/accuracy plot
-lr, --learnig_rate: Learnig Rate for the Model Training
-b, --batch_size: batch Size for model training
-e, --epochs: Epochs for Model Training

Example:

cd Liveness
python3 train.py --dataset data --batch_size 8 --epochs 50

Inference

Inference your Custom Liveness Model

Args

-m, --model: path to trained Liveness model
-i, --source: source - Video path or camera-id
-c, --conf: min prediction conf (0<conf<1)

Example:

cd Liveness
python3 inference.py --source 0 --conf 0.8

To Exit Window - Press Q-Key