Blog‎ > ‎

Deep Learning for Facial Expression Analysis

posted Jun 12, 2018, 5:22 PM by MUHAMMAD MUN`IM AHMAD ZABIDI   [ updated Jun 12, 2018, 5:33 PM ]
Machines can be taught to recognize expression using deep learning.

Example convolutional neural network (CNN) for facial expression analysis.

Deep learning consists of the training phase and inference/deployment phase.

During the training phase, CPUs and GPUs are used to find the best network architecture. This is lengthy process of finding the best number neural network layes, number of nodes per layer and the weights for each node. Typically, packages such as TensorFlow, Caffe or Theano are used. Matlab? not much.

CNN training phase.

After the architecture is defined, the coefficients (graphs and weights) derived from the training can be used for inference on an embedded system. This part has been successfully implemented by my Master's student on a Raspberry Pi.

CNN Inference phase.

Images: Synopsis