FacialRecognition

Facial Expression Recognition with TensorFlow

Introduction

• Facial Expression Recognition with Deep-Learning.
• Implementation CNN(Convolutional Neural Network) with TensorFlow 1.4.

Codes

• Test_Images : Directory of images for testing model.
• Train_Images : Directory of images for traning neural-network.
• collect_images.py : Collect face images from Bing and Google.
• convert_images.py : Convert images files(*.jpg, *.jpeg, .png) to dataset file(.bin).
• dataset.py : Dataset class for training or testing neural-network.
• cnn.py : Create CNN and train them or classify images.

Example of Run Codes

• Convert images to dataset

>>> import convert_images as ci

>>> ci.IMAGES_DIR = './Train_Images'
>>> ci.main('./train.bin', shuffle=True)

• Train CNN and save model

>>> from cnn import Cnn

>>> my_cnn = Cnn()

>>> my_cnn.set_device('gpu')
>>> my_cnn.set_epoch(1000)
>>> my_cnn.set_batch_size(100)

>>> my_cnn.train('./train.bin', './CNN_Models/model')

>>> del my_cnn

• Evaluate saved model

>>> from cnn import Cnn

>>> my_cnn = Cnn()

>>> my_cnn.set_device('gpu')
>>> my_cnn.set_batch_size(100)

>>> result = my_cnn.eval('./test.bin', './CNN_Models/model')
>>> print(result)

>>> del my_cnn

• Classify label of new images

>>> from cnn import Cnn

>>> my_cnn = Cnn()

>>> my_cnn.set_device('gpu')

>>> result = my_cnn.query('./new_image.jpg', './CNN_Models/model')
>>> print(result)

>>> del my_cnn

Example Images

• Angry

• Disgust

• Fear

• Happy

• Sad

• Surprise

• Neutral

References

• TensorFlow API r1.4