Emotion Detection from Facial Images

This project demonstrates how to train and evaluate a Convolutional Neural Network (CNN) to classify facial expressions using the CK+ dataset. It includes image preprocessing, model training, evaluation, and testing on unseen data.

Dataset

The dataset used for this project can be downloaded from Kaggle CK+ Dataset We use the CK+ (Extended Cohn-Kanade) dataset, which includes labeled facial expression images for emotions such as:

• Anger
• Contempt
• Disgust
• Fear
• Happy
• Sadness
• Surprise

Features

• End-to-end training pipeline using PyTorch
• Random image testing from dataset
• Confusion matrix and classification report
• Model saving for deployment
• Visualization of misclassified examples

Model Architecture

EmotionCNN

• (net): Sequential
• (0): Conv2d(1, 32, kernel_size=3, padding=1)
• (1): ReLU()
• (2): MaxPool2d(2)
• (3): Conv2d(32, 64, kernel_size=3, padding=1)
• (4): ReLU()
• (5): MaxPool2d(2)
• (6): Flatten()
• (7): Linear(9216, 128)
• (8): ReLU()
• (9): Linear(128, 7)

Results

• Validation Accuracy: 98.98% (on the CK+ subset used)
• Detailed precision, recall, and F1-scores per class

Requirements

• Python 3.10+
• PyTorch
• Torchvision
• Matplotlib
• Seaborn
• Scikit-learn

How to Run

1. Clone the repository
2. Download the CK+ dataset and place it in the expected folder
3. Open the Emotion_Detection_CKplus.ipynb notebook
4. Run all cells step-by-step

Model Saving

The model is saved as emotion_cnn.pth for reuse or deployment.

Project Screenshots

• Accuracy
• Confusion Matrix
• Test Results
• Misclassified example