A webcam that reads seven emotions.
A Flask app that reads seven emotions from a face, live from the webcam.
- Role
- Model and web app
- When
- Mar – Apr 2025
- Built with
- TensorFlow / Keras · OpenCV · Flask · pandas
- Where
- EdTechSociety · minor project
7Emotions
28,709Training faces
~62%Val. accuracy
The problem
Emotion demos usually stop at a notebook. This one had to work live, in a browser, on an ordinary webcam.
What I built
- The model: a 4-block CNN trained on FER-2013’s 28,709 faces across 7 emotions.
- The app: Flask, with photo upload or a live mode that reads a frame every second and answers with a colour and a wellbeing tip.
Results
About 62% validation accuracy across seven classes, on a hard, noisy benchmark. The architecture follows a well-known FER-2013 design; the live app is my own.
Next
Crop the face before classifying, save in Keras’s native format, and containerise.