The problem FacialEmotion_Detector solves

Human emotions play a critical role in communication, decision-making, and interaction. However, recognizing emotions accurately in real time is often challenging—especially for machines, automated systems, or even humans in fast-paced environments.

Challenges I ran into

Face Detection Accuracy

Challenge: The model sometimes struggled to detect faces in poor lighting or at different angles.

Solution: Used pre-trained models (Haar Cascades, Dlib, or MTCNN) and applied image preprocessing (grayscale conversion, histogram equalization) to improve detection accuracy.

Emotion Classification Errors

Challenge: The model often confused similar emotions (for example, fear vs. surprise, sad vs. neutral).

Solution: Improved training with a larger, more balanced dataset (FER2013, CK+) and applied data augmentation along with fine-tuning of the neural network.

Real-Time Processing Lag

Challenge: Running predictions on live video caused delays and high CPU/GPU usage.

Solution: Optimized the model with lightweight architectures (such as MobileNet) and reduced input image size to speed up inference.

Overfitting During Training

Challenge: The model performed well on training data but poorly on unseen data.

Solution: Applied dropout, early stopping, and regularization to improve generalization.

Integration with Webcam or UI

Challenge: Connecting the ML model with a real-time interface (OpenCV with Streamlit or Flask) caused compatibility issues.

Solution: Debugged step by step, used smaller test scripts to isolate errors, and ensured correct library versions.

Smart Contract Address(es)

0x9876543210abcdef9876543210abcdef98765432