Facial Expression Recognition Using Tensorflow

Used Convolutional neural networks (CNN) for facial expression recognition . The goal is to classify each facial image into one of the seven facial emotion categories considered .

Data :

We trained and tested our models on the data set from the Kaggle Facial Expression Recognition Challenge, which comprises 48-by-48-pixel grayscale images of human faces,each labeled with one of 7 emotion categories: anger, disgust, fear, happiness, sadness, surprise, and neutral .

Image set of 35,887 examples, with training-set : dev-set: test-set as 80 : 10 : 10 .

Dependencies

Python 2.7, Tensorflow, numpy .

Library Used:

• Tensorflow
• numpy

Train

To run the code -

1. Download FER2013 dataset from Kaggle Facial Expression Recognition Challenge and extract in the data/ folder.

2. After downloading dataset separate dataset into different csv for train ,dev and test. Run separate_data.py in data folder

   python separate_data.py
   

3. To train the model run train.py file

   python train.py
   

   Separate Model weights for each epoch is saved in model/ folder

4. Run evaluate.py to get accuracy on test data.
   
   "./model/model" + str(epochNumber) + ".ckpt" -> load specific epoch Model weight.
   Change this line in evaluate.py to choose which model weights should be loaded

  saver.restore(sess, "./model/model100.ckpt")