A Hackathon Project for BeachHacks 9.0 to detect user cortisol.

This project implements an interface that prompts a camera in order to take a picture. Using a Convolutional Neural Network (CNN), the program detects the stress level of the user. Taking into account time, weather, and facial expressions, the program rates the cortisol level of the user. Then, the user is given further advice and an assisstant chatbot in order to reduce these cortisol levels.

• CNN architecture for 48x48 grayscale images

• Regression output (continuous values instead of classes)

• Training with PyTorch

• MAE + Loss tracking

• Model saving and fine-tuning support

• Front end implementation with Streamlit

• Chatbot inquiry with uAgents

pip install -r requirements.txt

Dataset is from Kaggle: Fer2013. PIP and import kagglehub and run download_dataset.py to download the dataset.

• The dataset features ~30000 48x48 grayscaled images.

Models are saved within CNN_Model/saved_models Model: Convulutional Neural Network Current results:

• Classification model: 77% success rate
• Regression model: ~.18 MAE

Classification CNN:

• Loss: CrossEntropyLoss
• Optimizer: Adam
• Learning Rate: 1e-3
• Batch Size: 32

Regression CNN:

• Loss: SmoothL1Loss
• Optimizer: Adam
• Learning Rate: 1e-4
• Batch Size: 32

To continue training from a saved model, run CNN_Model/src/add_reg_training.py or CNN_Model/src/additional_training.py Make sure to tweak hyperparameters and model input/output.

Feel free to fork the repo and submit pull requests!

MIT License

Ryan Vo - ML Back End Omar Juarez - Streamlit Front End Fabien Duran - uAgent Full Stack