This program helps in identifying a person's emotion from live stream. The model inspiration was taken a research paper, where the authors published a efficient network for facial and gender identification (https://arxiv.org/pdf/1710.07557.pdf)
Just execute the below command from your terminal for installing all the required dependencies.
This program uses the network named exception (https://arxiv.org/pdf/1710.07557.pdf), which is a combination of several residual blocks. I trained the network for 100 epochs which took me 48 hours on the MAC with 16gb ram. However I strongly recommend training your model on Google Colab using GPU's, where it took me 45 minutes. 56 hours to 45 min :D
- For training your model, run
python training.py
. However I strongly recommend using model I trained it for you. - For using the model on a live stream, run
python RealtimeAnalysis.py
where you can capture the live stream from the webcam and the model inference will be displayed. - Notes: Use the gpu version of the model for the trained model
- Original Paper: https://arxiv.org/pdf/1710.07557.pdf
- Program Implementation of the published paper: https://github.com/oarriaga/face_classification
- Keras: https://keras.io/