Facial Expression Recognition using CNN

This repository contains a deep learning model built with PyTorch for facial expression recognition. The model classifies images of human faces into one of seven emotion categories: Angry, Disgust, Fear, Happy, Sad, Surprise, and Neutral.

---

Table of Contents

• Project Overview
• Features
• Dataset
• Installation
• Usage
• Training
• Evaluation
• Prediction
• Model Architecture
• Results
• License

---

Project Overview

Facial expression recognition is a fundamental task in human-computer interaction and affective computing. This project implements a Convolutional Neural Network (CNN) trained on a labeled dataset of facial images to classify emotions accurately.

---

Features

• Custom CNN architecture optimized for grayscale 48x48 pixel images.
• Data augmentation during training for improved generalization.
• Training and validation pipelines with accuracy and loss tracking.
• Model checkpointing for best validation accuracy.
• Inference script to predict emotion from an image URL.
• Modular and configurable codebase.

---

Dataset

The dataset consists of labeled facial images categorized into seven emotions:
Angry, Disgust, Fear, Happy, Sad, Surprise, Neutral.

The dataset should be organized with accompanying CSV files specifying image paths and labels. Images are expected to be grayscale and resized to 48x48 pixels during training and inference.

---

Installation

1. Clone this repository:

   git clone https://github.com/yourusername/face-expression-recognition.git
   cd face-expression-recognition

2. Create and activate a Python virtual environment (optional but recommended):

   python -m venv venv
   source venv/bin/activate  # Linux/macOS
   venv\Scripts\activate     # Windows

3. Install dependencies:

   pip install -r requirements.txt

---

Usage

Training the Model

Run the training script:

python train.py

The script will load the dataset, train the CNN model, and save the best performing model to the checkpoints/ directory.

Predicting Emotion from Image URL

Use the prediction script to classify an emotion from an image URL:

python predict.py --url "https://example.com/image.jpg"

The script outputs the predicted emotion label.

---

Training

• The model trains using CrossEntropyLoss and Adam optimizer.
• Training and validation data loaders use data augmentation and normalization.
• Training progress is displayed with loss and accuracy metrics per epoch.
• The best model based on validation accuracy is saved automatically.

---

Evaluation

Validation accuracy is printed after each epoch. Typical performance achieves approximately 50% accuracy on validation data, which can be improved with further tuning and data.

---

Model Architecture

The CNN model consists of:

• 3 convolutional layers with batch normalization and max pooling.
• Dropout layer to reduce overfitting.
• Fully connected layers mapping to 7 output emotion classes.

Input images are grayscale of size 48x48 pixels.

---

Results

Example metrics after training:

• Train Loss: 1.38
• Train Accuracy: 46.6%
• Validation Accuracy: 51.3%

---

License

This project is licensed under the MIT License - see the LICENSE file for details.

---