This repository contains working code used to train an emotion recognition model that can be used to automatically assess audience experience. The training code can be found in Emotion_Recognition_Module.ipynb. We also provide an implementation of the full experience assessment pipeline, which captures video from a webcam, identifies and tracks the people in the video, and assigns experience scores for each person found.

The training script requires the following dependencies

• CV2: 4.5.2
• PIL: 8.2.0
• torch: 1.8.1+cu111
• torchvision: 0.9.1+cu111
• matplotlib: 3.4.2
• sklearn: 0.24.2
• numpy: 1.20.3

In addition, the full pipeline makes use of darknet to perform detection using YOLOv4 Note that although we used CUDA, this is not a mandatory requirement.

After installing the dependencies, move the *.py files into the darknet directory and run python full_pipeline.py, or launch a jupyter server if you want to train the system.