Face Recognition Model - Gender Classification

A deep learning project implementing a binary face recognition system to classify faces by gender using transfer learning with VGG16 architecture.

📋 Project Overview

This project demonstrates the implementation of a CNN-based face recognition model that achieves high accuracy in gender classification. The model leverages transfer learning from VGG16 pre-trained on ImageNet and includes advanced data augmentation, performance monitoring, and comprehensive evaluation metrics.

🎯 Key Features

• Transfer Learning: Utilizes VGG16 architecture pre-trained on ImageNet
• Enhanced Data Augmentation: Random flips, rotations, zooms, and contrast adjustments
• Model Checkpointing: Automatic saving of best-performing model based on validation loss
• Comprehensive Evaluation: Classification reports, confusion matrix, and visual analysis
• Performance Visualization: Training/validation metrics tracking and prediction display

🛠️ Tech Stack

• Framework: TensorFlow/Keras
• Base Model: VGG16 (Pre-trained on ImageNet)
• Libraries:
  • TensorFlow Datasets
  • NumPy
  • Matplotlib
  • Scikit-learn
  • Pickle (for model persistence)
• Image Processing: TensorFlow image preprocessing
• Data Augmentation: Keras data augmentation layers

📊 Model Architecture

• Base: VGG16 (frozen layers)
• Custom Layers:
  • Global Average Pooling
  • Dense layers (256, 128 units) with ReLU activation
  • Dropout (0.5) for regularization
  • Output layer (2 units) with Softmax activation

🚀 Setup Instructions

1. Clone the repository

git clone https://github.com/KRT2002/face-classification-deep-learning.git
cd face-classification-deep-learning

2. Create virtual environment

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install dependencies

pip install -r requirements.txt

4. Run the notebook Open and execute the Jupyter notebook to train and evaluate the model.

📈 Results

The model demonstrates strong performance across both classes:

• Overall Accuracy: ~95%+
• Precision & Recall: High scores for both genders
• Model Generalization: Validation metrics closely track training metrics, indicating good generalization without overfitting

For detailed results, metrics, and visualizations, refer to DOCUMENTATION.md.

📁 Project Structure

├── Face Recognition model.docx    # Original documentation
├── DOCUMENTATION.md               # Detailed implementation guide
├── notebook.ipynb                 # Implementation notebook
├── requirements.txt               # Python dependencies
└── README.md                      # This file

💡 Implementation Highlights

1. Data Preprocessing: Images resized to 224x224, normalized and batched
2. Class Balance Handling: Dataset analysis shows balanced distribution (minority class >30%)
3. Callbacks: ModelCheckpoint monitors validation loss to save best model
4. Evaluation: Multi-metric analysis including confusion matrix and classification report

📝 Usage

Refer to the implementation notebook for step-by-step code execution. The notebook includes:

• Dataset loading and exploration
• Data augmentation pipeline
• Model building and training
• Performance evaluation
• Prediction on custom images

🔍 Key Learnings

• Transfer learning effectiveness for face recognition tasks
• Importance of data augmentation for model robustness
• Model checkpoint strategies for optimal model selection
• Comprehensive evaluation beyond simple accuracy metrics

---

This project was developed to demonstrate deep learning capabilities in computer vision and binary classification tasks.