Emojify: Real-Time Facial Emotion Recognition Using CNN — CNN training/inference project (Mar 2025 - Aug 2025)
Designed and trained a CNN for real-time facial emotion recognition using the FER-2013 dataset. The experience involved preparing and using labeled facial emotion data to train a classification model and then evaluating it with standard performance visualizations. The model was integrated for live inference using computer vision and a UI. • Trained CNN on FER-2013 for emotion classification (93.5% accuracy). • Generated confusion matrices and accuracy comparison charts for evaluation. • Benchmarked against SVM, Random Forest, and RNN baselines. • Integrated model with OpenCV and Streamlit for real-time detection.