amineHorseman/facial-expression-recognition-using-cnn

Combines raw pixel data with dlib face landmarks and HOG features as CNN inputs, demonstrating that auxiliary features improve accuracy on the challenging FER2013 dataset. Implements two CNN architectures with batch normalization, achieving 75.1% on 5-expression classification and supporting hyperparameter optimization via Hyperopt. Includes preprocessing pipeline for face detection via OpenCV and real-time inference from video/camera streams.

511 stars. No commits in the last 6 months.