DeepFace Face Recognition Tool

A robust and accurate face verification system that leverages state-of-the-art deep learning models to compare and verify faces across different images. This tool uses the DeepFace library and provides multiple models, detectors, and visualization options.

Features

• Multiple Face Recognition Models:

  • ArcFace (default) - One of the most accurate face recognition models
  • VGG-Face, Facenet, OpenFace, DeepID, and more
  • Ensemble option that uses multiple models and votes on the result

• Advanced Face Detection Backends:

  • RetinaFace (default) - State-of-the-art face detector
  • MTCNN, SSD, MediaPipe, Dlib, and OpenCV options
  • Automatic fallbacks if detection fails

• Comprehensive Analysis:

  • Similarity scores and confidence levels
  • Distance metrics with proper thresholds
  • Detailed results with multiple metrics
  • Visual comparison with automated decision

• Flexible Options:

  • Choose distance metrics (cosine, euclidean)
  • Toggle face detection enforcement
  • Create visualizations of results
  • Debug options for troubleshooting

Installation

Prerequisites

• Python 3.6+
• pip package manager

Quick Install

# Clone this repository
git clone https://github.com/subhashdasyam/compare-faces.git
cd compare-faces

# Install dependencies
pip install -r requirements.txt

The tool will automatically attempt to install DeepFace if not found on your system.

Manual Dependencies Installation

If you prefer to install dependencies manually:

pip install deepface opencv-python matplotlib numpy

Usage

Basic Usage

python deepface_recognition.py image1.jpg image2.jpg

This will:

1. Compare the faces in the two images
2. Generate a JSON report with similarity scores and confidence
3. Create a visualization image in the output directory

Advanced Usage

Use a specific face recognition model:

python deepface_recognition.py image1.jpg image2.jpg --model ArcFace

Available models: VGG-Face, Facenet, Facenet512, OpenFace, DeepFace, DeepID, ArcFace, Dlib, SFace, ensemble

Use a different face detector:

python deepface_recognition.py image1.jpg image2.jpg --detector retinaface

Available detectors: opencv, ssd, dlib, mtcnn, retinaface, mediapipe

Change the distance metric:

python deepface_recognition.py image1.jpg image2.jpg --metric cosine

Available metrics: cosine, euclidean, euclidean_l2

Handle challenging images:

python deepface_recognition.py image1.jpg image2.jpg --no-enforce-detection

Use ensemble of multiple models (most accurate):

python deepface_recognition.py image1.jpg image2.jpg --model ensemble

Disable visualization:

python deepface_recognition.py image1.jpg image2.jpg --no-visualize

Enable debug output:

python deepface_recognition.py image1.jpg image2.jpg --debug

Command Line Options

Option	Description
image1	Path to first image (required)
image2	Path to second image (required)
--model	Face recognition model to use (default: ArcFace)
--detector	Face detector backend (default: retinaface)
--metric	Distance metric (default: cosine)
--no-enforce-detection	Don't enforce face detection (helpful for challenging images)
--no-visualize	Don't create visualization
--debug	Enable debug output

Example Output

{
  "match": false,
  "distance": 0.5623421072959899,
  "threshold": 0.40000000000000036,
  "similarity": 0.4376578927040101,
  "confidence": "low",
  "model": "ArcFace",
  "detector": "retinaface",
  "distance_metric": "cosine",
  "processing_time": 2.345918893814087,
  "message": "Match not found with low confidence (0.4377)"
}

Output Files

The tool generates the following files in the output directory:

• verification_result.png: Visual comparison of the two images with matching results
• original_img1.jpg and original_img2.jpg: Original input images
• face1.jpg and face2.jpg: Extracted face regions (if detection succeeds)

Troubleshooting

Face detection fails

If face detection fails, try the following:

• Use --no-enforce-detection flag
• Try a different detector with --detector mtcnn or --detector ssd
• Check if the image is too low resolution or the face is too small

DeepFace installation issues

If you encounter issues with automatic DeepFace installation:

1. Try manual installation: pip install deepface
2. Check Python version (3.6+ required)
3. Ensure you have proper GPU drivers if using GPU acceleration

Out of memory errors

If you get out of memory errors:

1. Try a lighter model: --model VGG-Face or --model OpenFace
2. Use a less resource-intensive detector: --detector opencv

Requirements

• Python 3.6+
• DeepFace
• OpenCV
• Matplotlib
• NumPy

License

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

Acknowledgments

• DeepFace library by Sefik Ilkin Serengil
• Face recognition models: ArcFace, VGG-Face, FaceNet, etc.