Facial Classifier

Computer vision web application, built to predict the age, race, and gender of all individuals present in an image

• Achieves an average 85% accuracy across all three dependent variables
• CNN model built and trained in PyTorch
• Developed with a Flask Python backend and bootstrap frontend

Testing it out on myself!

Running the website locally

This facial classifier runs on Python 3.7

Start off by cloning the repo:
git clone https://github.com/danielzgsilva

Navigate to the project's root and install dependencies like so:
pip install -r requirements.txt

Run the app with:
python app.py

The project will then serve locally on port 5555:
http://localhost:5555/

Model details

The underlying Convolutional Neural Network uses a pretrained Squeeze and Excitation Network (SENet), trained on VGGFace2

• Cao, Qiong, et al. "Vggface2: A dataset for recognising faces across pose and age." Automatic Face & Gesture Recognition (FG 2018), 2018 13th IEEE International Conference on. IEEE, 2018.

The model was fine tuned for classification of tightly cropped facial images using the UTKFace dataset

• https://susanqq.github.io/UTKFace/

Work in progress:

• Deploying the app through Heroku
• Combatting learned biases due to imbalanced datasets
• Adding the ability to take a picture through the app, rather than requiring an image upload
• Increasing performance in poor lighting