Face Recognision.

Face Recognision with VGG and Resnet model by using Transfer learning.

Requirements

• Keras
• Tensorflow
• Numpy
• Pandas
• Opencv2
• Matplotlib
• Keras - VGG Documentation for Keras VGG

Files & Folders

• faces - Faces of persons kept inside each Folder named with person's Name.
• Images - Images of person Kept inside Directory Named with person's Name.
• video_frames - Frames of person kept inside folder with persons name if video is used to collect Images.
• face_extrator.py - To seperate faces from photos.
• frame_capture.py - Use to seperate frames from individual videos.
• live_input.py - To test final model after training which take input from Webcam and recognise the persons in video
• model_train.py - To Train Model after seperating each persons faces and keeping them in Correct Folders.

Implementation

• Input from video

  • Place the persons video in Main directory and start frame_capture.py file on each Video
  • Code - '''Shell python frame_capture.py name_1.mp4 '''
  • This will place frames from each video in video_frames folders inside name_1 folder.

• Faces from Frames

  • You can take input from Video or collect individual photos of each person and keep inside images folder
  • Code - '''Shell python face_extractor.py video_frames/name_1 or python face_extractor.py images/name_1 '''
  • This will extract faces from photos and place them faces folder inside there corresponding Folders.

• Start Training -

  • Before starting Make sure the Directories are formed in correct order.
  • faces Folder have structure -> faces

  ----> name_1

  --------> file_of_name_1

  --------> file_of_name_1

  --------> file_of_name_1

  --------> file_of_name_1

  ----> name_2

  --------> file_of_name_2

  --------> file_of_name_2

  --------> file_of_name_2

  --------> file_of_name_2

  ----> name_3

  --------> file_of_name_3

  --------> file_of_name_3

  --------> file_of_name_3

  --------> file_of_name_3

  ...

  • Start the training with Command " python model_train.py resnet " or " python model_train.py vgg " to train model with either Resnet50 or VGG16 Model . VGG16 is smaller model and trains faster and requires less RAM Compare to Resnet50 model.
  • If everything is successful you will find named Model_VGGFace.h5 inside main directory.

• Test Model -

  • Start the video input from Webcam by command " python live_input.py "
  • If everything is Successful you will find Rectangle around faces with name of person written over it.

Note

• You can change values inside file face_extractor.py file and live_input.py file and play with scaleFactor , minNeighbors of HaarCascade .These values work fine for me.
• Some Manual Assist may require after face_extration as some wrong files may come along with faces.

Related Files

• https://github.com/rcmalli/keras-vggface