https://github.com/tahirzia-1/facial-emotion-recognition

An AI model to Recognize Facial Emotions from the (Universal) 7 classes of emotions.
https://github.com/tahirzia-1/facial-emotion-recognition

ai classes cnn cnn-classification cnn-keras facial-expression-recognition facial-recognition ipynb-jupyter-notebook

Last synced: 8 months ago
JSON representation

An AI model to Recognize Facial Emotions from the (Universal) 7 classes of emotions.

• Host: GitHub
• URL: https://github.com/tahirzia-1/facial-emotion-recognition
• Owner: TahirZia-1
• Created: 2024-05-08T15:52:21.000Z (over 1 year ago)
• Default Branch: main
• Last Pushed: 2025-02-26T18:06:03.000Z (8 months ago)
• Last Synced: 2025-02-26T18:46:12.174Z (8 months ago)
• Topics: ai, classes, cnn, cnn-classification, cnn-keras, facial-expression-recognition, facial-recognition, ipynb-jupyter-notebook
• Language: Jupyter Notebook
• Homepage:
• Size: 84 KB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# Facial Expression Recognition with EfficientNetB2

This repository implements a facial expression recognition system using the EfficientNetB2 model from TensorFlow/Keras. The project classifies facial images into seven emotion categories (angry, disgust, fear, happy, neutral, sad, surprise) using a dataset with 28,821 training and 7,066 validation images. The trained model leverages transfer learning, data augmentation, and class weighting to handle imbalanced data, and it’s saved for future inference.

## Overview

The goal is to classify facial expressions from RGB images (96x96) into seven emotion classes. The dataset is preprocessed with data augmentation and one-hot encoding, and the model is trained with class weights to address class imbalance. The EfficientNetB2 backbone is fine-tuned with additional layers for improved performance.

### Key Features

- **Dataset**: 28,821 training and 7,066 validation images across 7 classes.

- **Model**: EfficientNetB2 with custom Conv2D, GlobalAveragePooling2D, Dense, and Dropout layers.

- **Preprocessing**: Resizing, random flips, rotations, and zooms for augmentation.

- **Class Balancing**: Class weights to mitigate imbalance (e.g., "disgust" has fewer samples).

- **Evaluation**: Accuracy, precision, and recall metrics.

## Implementation

### Dataset

- **Structure**: 

  - `images/train/`: 28,821 images (7 subfolders for each class).

  - `images/validation/`: 7,066 images (7 subfolders).

- **Classes**: Angry (0), Disgust (1), Fear (2), Happy (3), Neutral (4), Sad (5), Surprise (6).

- **Preprocessing**: Images resized to 96x96, normalized, and augmented.

### Model Architecture

- **Input**: 96x96x3 (RGB images).

- **Backbone**: EfficientNetB2 (pre-trained, no top).

- **Custom Layers**:

  - Conv2D: 128 filters, 3x3 kernel, ReLU activation.

  - GlobalAveragePooling2D.

  - Dense: 128 units, ReLU activation.

  - Dropout: 0.3.

  - Dense: 7 units, softmax activation (output).

- **Parameters**: ~9.41M total, ~9.34M trainable.

### Training

- **Optimizer**: Adam (learning rate: 0.001).

- **Loss**: Categorical Crossentropy.

- **Metrics**: Accuracy, Precision, Recall.

- **Epochs**: 8.

- **Batch Size**: 64.

- **Class Weights**: Computed to balance class distribution (e.g., `{0: 1.79, 1: 16.43, ...}`).

### Results

- **Training**:

  - Best Accuracy: 85.66% (Epoch 1).

  - Precision: ~87%, Recall: ~83%.

- **Validation**:

  - Accuracy: ~1.58% (Epoch 8).

  - Loss: 2.1885, Precision: 0.0, Recall: 0.0 (indicating overfitting).

- **Note**: Validation performance suggests overfitting; further tuning is needed.

*Class Distribution Plot*:  

![Class Distribution](images/class_distribution.png)  

(Saved from `facial_expression_recognition.ipynb` as `images/class_distribution.png`.)

## Installation

### Prerequisites

- Python 3.x

- Required libraries:

  - TensorFlow (`pip install tensorflow`)

  - NumPy (`pip install numpy`)

  - Pandas (`pip install pandas`)

  - Matplotlib (`pip install matplotlib`)

  - Seaborn (`pip install seaborn`)

  - OpenCV (`pip install opencv-python`)

  - Scikit-learn (`pip install scikit-learn`)

### Setup

1. **Clone the Repository**:

   ```bash

   git clone https://github.com/TahirZia-1/facial-expression-recognition.git

   cd facial-expression-recognition