Development environment setup (Windows 8.1):

• Install Python 3.5.5
• Install Tensorflow version 1.8.0 (CPU)
• [OR] Install Tensorflow version 1.8.0 (GPU)
  • CUDA® Toolkit 9.0
  • cuDNN v7.0
• External Packages Required:
  • scipy
  • scikit-learn
  • opencv-python
  • H5py
  • numpy
  • matplotlib
  • Pillow
  • Requests
  • pickle
  • psutil
  • shutil

Capturing Images:

• Run capture_im.py [Remove meta.csv and WebcamCap.txt before proceeding to train/classify]
• Using:

  filepath = ' ..\datasets\dataset_name\raw\'
  fileName = ' ..\datasets\dataset_name\raw'+'\WebcamCap.txt'

Organization of Dataset:

• The raw images are organized as: datasets\dataset_name\raw\subject_name

  gtfd\raw\person01
        ..\person02
  

• Each subject_name folder contains the images.

  gtfd\raw\person1\image0
                ..\image1
  

Download and save facenet pre-trained models as: ..\models\facenet\20180408-102900\

• 20180408-102900
• 20180402-114759

Alignment of dataset using MTCNN:

• SET PYTHONPATH=%PYTHONPATH%;'..\facenet\src'
• Run align_dataset_mtcnn.py

    python align\align_dataset_mtcnn.py 
    ..\facenet\datasets\dataset_name\raw        
    ..\facenet\datasets\dataset_name\aligned
    --image_size 182  
    --margin 44 
    --random_order
    --gpu_memory_fraction 0.85 
  

Split Dataset to Test and Train:

• Run test_train_split.py
• Set the ratio of test data and directory:

  percent = 0.8
  add = ‘directory of aligned dataset’

Train a classifier on the training data:

• SET PYTHONPATH=%PYTHONPATH%;'..\facenet'
• Run classifier_svm.py or classifier_mlp.py

  python src\classifier_mlp.py
  TRAIN
  datasets\dataset_name\aligned\Train models\facenet\20180408-102900\20180408-102900.pb       
  models\my_classifier.pkl 
  --batch_size 500
  

Test the classifier on testing data:

• SET PYTHONPATH=%PYTHONPATH%;'..\facenet'
• Run classifier_svm.py or classifier_mlp.py

  python src\classifier_mlp.py
  CLASSIFY
  datasets\dataset_name\aligned\Test models\facenet\20180408-102900\20180408-102900.pb       
  models\my_classifier.pkl 
  --batch_size 500
  

To run Real-time face recognition instead:

• SET PYTHONPATH=%PYTHONPATH%;'..\facenet\realtime_facenet'
• Run start_realtime.py
• Update classifier file name, model directory and dataset directory in start_realtime.py