Face Verification with MTCNN and InceptionResnetV1

A deep learning-based face verification system using MTCNN for face detection and InceptionResnetV1 for face recognition, fine-tuned on custom dataset.

📊 Architecture Overview

graph LR
    A[Input Image] --> B[MTCNN]
    B --> C[Face Detection]
    C --> D[Face Alignment]
    D --> E[InceptionResnetV1]
    E --> F[Face Embeddings]
    F --> G[Verification Result]
    
    style A fill:#f9f,stroke:#333,stroke-width:2px
    style G fill:#bbf,stroke:#333,stroke-width:2px

Loading

🛠️ Technical Details

Requirements

• facenet_pytorch
• torch
• torchvision
• numpy
• opencv-python
• tensorboard

Model Architecture

graph TD
    A[Input Layer] --> B[MTCNN]
    B --> C[InceptionResnetV1]
    C --> D[FC Layer]
    D --> E[Output Layer]

Loading

🔧 Implementation Details

Data Processing

• Face detection using MTCNN
• Image resizing to 160x160 pixels
• Face alignment and normalization
• Data augmentation techniques

Training Configuration

• Batch Size: 16
• Epochs: 120
• Optimizer: Adam
• Learning Rate: 0.001
• Loss Function: CrossEntropyLoss
• Learning Rate Scheduler: MultiStepLR

🔍 Usage

1. Data Collection:

# Collect and organize face images in the following structure:
/dataset
    /Person1
        image1.jpg
        image2.jpg
    /Person2
        image1.jpg
        image2.jpg

2. Training:

# Run the training script
python training_mtcnn.py

3. Model Inference:

# Load the trained model
model = InceptionResnetV1(pretrained='vggface2')
model.load_state_dict(torch.load('Face_Verification_v4.pth'))

🎯 Features

• Robust face detection using MTCNN
• Custom data augmentation pipeline
• Fine-tuned InceptionResnetV1 model
• Learning rate scheduling
• Training and validation visualization
• Cross-entropy loss optimization

📊 Results

The model achieves:

• Training Accuracy: ~95%
• Validation Accuracy: ~93%
• Real-time inference capability
• Robust face verification performance

🔗 Project Structure

graph TD
    A[Project Root] --> B[training_mtcnn.py]
    A --> D[dataset/]
    A --> E[models/]
    D --> F[Person1/]
    D --> G[Person2/]
    E --> H[Face_Verification_v4.pth]

Loading

🤝 Contributing

Feel free to open issues and pull requests for:

• Bug fixes
• New features
• Documentation improvements
• Performance optimizations

📝 License

This project is licensed under the Appache 2.0 License - see the LICENSE file for details.

🙏 Acknowledgments

• FaceNet PyTorch implementation
• MTCNN paper and implementation
• Labelled Faces in the Wild (LFW) Dataset