Face Recognition Fast

A developer-friendly, multi-method face recognition system featuring:

• OpenCV (Haar Cascade),
• MTCNN,
• face_recognition (dlib-based),
• MobileNet (TensorFlow/Keras),
• ArcFace (InsightFace framework),
• DeepFace (multi-backend wrapper).

Compare speed and memory usage of these approaches on your own images, or just pick a single method to run.

Table of Contents

• Features
• Setup & Installation
• Usage
  • Single Method
  • Benchmark Mode
  • Example Output
• Project Structure
• Contributing
• License

---

Features

1. Multiple Face Detection & Embedding Methods

   • OpenCV for detection, with face_recognition embeddings
   • MTCNN for detection (TensorFlow-based)
   • face_recognition for both detection + embedding
   • MobileNet (MTCNN detection + MobileNetV2 embedding)
   • ArcFace (RetinaFace detection + ArcFace embedding)
   • DeepFace (multi-backend with Facenet, VGG-Face, ArcFace, etc.)

2. Benchmark Mode

   • Runs all six methods in a single pass and reports:
     • Execution time
     • Memory usage difference (MemΔ)
     • L2 distance
     • Cosine similarity

3. Local (Lazy) Imports

   • TensorFlow only loads if you actually use MTCNN, MobileNet, or DeepFace.
   • This keeps overhead and logs minimal for non-TF methods.

4. Easy CLI

   • Just provide two image paths (reference & test images) plus a method (or --benchmark).

---

Setup & Installation

1. Clone the Repository

git clone https://github.com/pescheckit/face-recognition-fast.git
cd face-recognition-fast

2. Create a Virtual Environment

To keep things clean and self-contained, create a Python virtual environment:

python3 -m venv .venv
source .venv/bin/activate

On Windows, use:

.venv\Scripts\activate

3. Install Dependencies

Install all required libraries from requirements.txt:

pip install --upgrade pip
pip install -r requirements.txt

---

Usage

From the face-recognition-fast folder:

python main.py --help

Single Method

Choose exactly one of the six methods:

• opencv
• mtcnn
• facerecognition
• mobilenet
• arcface
• deepface

Example:

python main.py images/id1.png images/front1.png --method arcface

Benchmark Mode

Compare all six methods at once:

python main.py images/id1.png images/front1.png --benchmark

Example Output

===== BENCHMARK RESULTS =====
+-----------------+----------+-----------+--------------+----------+------------+--------------+
| Method          | Status   |   Time(s) |   Memory(MB) |   CPU(%) |   Distance | Similarity   |
+=================+==========+===========+==============+==========+============+==============+
| opencv          | OK       |    1.0056 |       212.3  |    27.09 |     0.424  | 96.23%       |
+-----------------+----------+-----------+--------------+----------+------------+--------------+
| mtcnn           | OK       |    2.7672 |       562.93 |    11.34 |     0.4499 | 95.60%       |
+-----------------+----------+-----------+--------------+----------+------------+--------------+
| facerecognition | OK       |    0.7948 |        20.86 |    20.12 |     0.429  | 96.18%       |
+-----------------+----------+-----------+--------------+----------+------------+--------------+
| mobilenet       | OK       |    1.8644 |        50.74 |     8.11 |    18.7707 | 65.85%       |
+-----------------+----------+-----------+--------------+----------+------------+--------------+
| arcface         | OK       |    1.7279 |       688.3  |    18.55 |     2.1713 | 94.79%       |
+-----------------+----------+-----------+--------------+----------+------------+--------------+
| deepface        | OK       |    2.3077 |        30.41 |    13.59 |  2986.94   | 93.25%       |
+-----------------+----------+-----------+--------------+----------+------------+--------------+

---

Project Structure

face-recognition-fast/
├── main.py
├── src/
│   ├── face_detection.py
│   ├── face_embedding.py
│   ├── compare_faces.py
│   └── utils.py
├── images/
│   ├── id1.png
│   └── front1.png
├── requirements.txt
└── README.md

---

Contributing

1. Fork the project on GitHub.
2. Create a new branch:

   git checkout -b my-feature

3. Make your changes and commit:

   git commit -am "Add new feature"

4. Push to your branch:

   git push origin my-feature

5. Open a Pull Request on GitHub and describe your changes.

---

License

This project is distributed under the MIT License.