ml-lab / facial-expression-project

Can we predict facial expressions?

Facial Expression Classification Web App

Chris Green, March 2015 😄 😦 😮

##Contents

• TL;DR
• Overview
• Web App
• Model
• Installation

##TL;DR This Django project classifies a person's face as either happy, sad, or surprised. The live web-app can be found [here][1]. The convolutional neural network training code and documentation can be found in [this repo][2].

##Overview Can we teach a machine to recognize human emotions? That was the question I wanted to tackle in my capstone project for [Zipfian Academy's][4] winter 2015 data science cohort. Luckily I found a large dataset of labeled grayscale images of faces broken into 7 different classes [here][5]. For modeling I choose to experiment with [convolutional neural networks][6], a type of deep learning model especially suited for image recognition. Training using all 7 labels proved challenging so I reduced the problem down to 3 expression: Happy, Sad, and Surprised. Even with this reduction in scope we still had ~19,000 images to work with. The final model was able to acheive 75% accuracy. ##Web App This [Django][3] project creates a site to showcase the [facial expression classifier][2] package by loading in images from the the web and classifying them. To use the site simply add a url of an image with a face in it and see what you get. Be sure to make sure it is a valid and accessible image. For example submitting this [url](http://images.sodahead.com/polls/003482687/2111968424_sad_mitt_romney_face_answer_1_xlarge.jpeg) gives:

##Model A detailed overview of the facial expression classification package used in this app can be found [here][2]. ##Installation To install and run this app you will need register for [GraphLab Create][7] (Don't worry...it's free). You also need to install [OpenCV][8]. After installing GraphLab and OpenCV run the following bash commands: ```bash git clone https://github.com/cmgreen210/facial-expression-project cd facial-expression-project pip install -r requirements.txt python manage.py migrate python manage.py runserver open http://127.0.0.1:8000/ ```