Emotion-detection

Why we included this project

For a compact, readable example of real-time facial emotion recognition, this is a handy place to start. It chains a Haar cascade face detector to a small convolutional network trained on FER-2013, resizing each detected face to 48x48 before classifying it into one of seven emotions, from angry to surprised. Everything lives in a single script with train and display modes, so you can either train the model yourself and reproduce the roughly 63% test accuracy, or download the pre-trained weights and run live webcam predictions right away. The repo also includes the code used to convert the raw FER-2013 CSV into PNG images, which saves beginners a chunk of fiddly data prep. It is a teaching-oriented project built around a simple CNN rather than a production inference service, but that is exactly what makes it a clear starting point for classroom work or hobby experiments.