travistangvh / emotion-detection-in-real-time

Face Classification

The following repository is a real-time face detection and emotion classification model.

The face detection is powered by MTCNN and openCV. The emotion classification model is a built on an CNN architecture called VGGFace with weights trained on the fer2013 dataset.

The Model

The model is trained on a CNN architecture called VGGFace.

Instructions on getting started

To run the demo.

• Clone this commit to your local machine using git clone https://github.com/travistangvh/emotion-detection-in-real-time.git

• Install these dependencies with pip install pip install -r ../REQUIREMENTS.txt

• Download pretrained model and weight trained_vggface.h5 from here.

• Place trained_vggface.h5 into ../datasets/trained_models/.

• Run emotion_webcam_demo.py using python3 emotion_webcam_demo.py

To train previous/new models for emotion classification:

• Download the fer2013.tar.gz file from here

• Move the downloaded file to the ../datasets/raw/ directory inside this repository.

• Untar the file: tar -xzf fer2013.tar

• Ensure that the file ../datasets/raw/fer2013.csv exists

• Run the training_emotion_classification.py file python3 training_emotion_classifier.py

Citations

• Deep Face Recognition by Parkhi et. al.