A goal was set
Myself, Susan Kilean Hwang, Susan Hao, and Zhimin Chen decided to build a computer vision system which would be able to label the displayed emotion on a given human face. With consultation from Frank Cleary we decided a system based on live webcam video had more potential future use.
As a team we took this idea from the whiteboard to the web. We trawled the web for data, built and tested several models quickly selecting a transfer model based off of the VGG-16 architecture. Using cv2 greatly reduced the amount of back-end effort we would have had to put into image preparation for our model in the on-line pipeline.
We successfully developed a Flask based web app which exceeded our expectations - in real time it was able to identify one or multiple human faces and correctly name the emotion present on those faces even as the faces transitioned their emotional state.
With my background in engineering optimization I'm proud of what I see as my key contribution, saving the team a little over a day in hyperparameter search time via a simple linear minimization approach. I enjoyed pair-coding and helping to build the various components of our model, user interface, etc., but saving the team time is my own contribution I can smile about.