Facial Recognition Pipeline with Deep Learning in TensorFlow

Table of Contents

• Overview
• Prerequisites
• Project Structure
• Setup and Installation
• Data Preparation
• Preprocessing
• Generating Embeddings
• Training the Classifier
• Evaluation
• References

Overview

This project implements a facial recognition pipeline using TensorFlow, Dlib, and Docker. The pipeline encompasses:

1. Preprocessing Images: Detecting and aligning faces in images using Dlib to ensure consistency.
2. Generating Facial Embeddings: Utilizing a pre-trained convolutional neural network to extract 128-dimensional feature vectors (embeddings) representing each face.
3. Training an SVM Classifier: Using the embeddings to train an SVM classifier that can distinguish between different identities.

Prerequisites

Before running the project, ensure you have the following:

• Basic understanding of Linear Algebra
• Basic understanding of Convolutional Neural Networks
• Basic knowledge of TensorFlow

Project Structure

├── Dockerfile
├── etc
│   ├── 20170511-185253
│       ├── 20170511-185253.pb
├── data
├── medium_facenet_tutorial
│   ├── align_dlib.py
│   ├── download_and_extract_model.py
│   ├── __init__.py
│   ├── lfw_input.py
│   ├── preprocess.py
│   ├── shape_predictor_68_face_landmarks.dat
│   └── train_classifier.py
├── requirements.txt

Setup and Installation

1. Clone the Repository:

   git clone https://github.com/YourUsername/your-repo-name.git
   cd your-repo-name

2. Install Docker:

   Ensure Docker is installed on your system. You can install it by running:

   curl -sSL https://get.docker.com/ | sh

3. Build the Docker Image:

   docker build -t facial_recognition_pipeline -f Dockerfile .

   Note: Building the image may take several minutes, depending on your hardware.

Data Preparation

1. Download the LFW Dataset:

   The Labeled Faces in the Wild (LFW) dataset will be used for training. Download and extract it as follows:

   curl -O http://vis-www.cs.umass.edu/lfw/lfw.tgz
   tar -xzvf lfw.tgz

   Note: Verify the integrity of the downloaded dataset by comparing its checksum with the one provided on the official website to avoid using corrupted files.

   Ensure the dataset is structured as:

   data/
   ├── person1
   │   ├── image1.jpg
   │   └── image2.jpg
   ├── person2
   │   ├── image1.jpg
   │   └── image2.jpg
   ...
   

Preprocessing

1. Download Dlib's Face Landmark Predictor:

   curl -O http://dlib.net/files/shape_predictor_68_face_landmarks.dat.bz2
   bzip2 -d shape_predictor_68_face_landmarks.dat.bz2

2. Run the Preprocessing Script:

   Use the provided script to detect, align, and crop faces from the dataset:

   docker run -v $PWD:/facial_recognition_pipeline \
   -e PYTHONPATH=$PYTHONPATH:/facial_recognition_pipeline \
   -it facial_recognition_pipeline python3 /facial_recognition_pipeline/medium_facenet_tutorial/preprocess.py \
   --input-dir /facial_recognition_pipeline/data \
   --output-dir /facial_recognition_pipeline/output/intermediate \
   --crop-dim 180

Generating Embeddings

1. Download the Pre-trained Model:

   Execute the script to download the pre-trained FaceNet model:

   docker run -v $PWD:/facial_recognition_pipeline \
   -e PYTHONPATH=$PYTHONPATH:/facial_recognition_pipeline \
   -it facial_recognition_pipeline python3 /facial_recognition_pipeline/medium_facenet_tutorial/download_and_extract_model.py \
   --model-dir /facial_recognition_pipeline/etc

2. Generate Embeddings:

   Run the script to produce embeddings for each aligned face image.

Training the Classifier

1. Train the SVM Classifier:

   Train the classifier using the generated embeddings:

   docker run -v $PWD:/facial_recognition_pipeline \
   -e PYTHONPATH=$PYTHONPATH:/facial_recognition_pipeline \
   -it facial_recognition_pipeline \
   python3 /facial_recognition_pipeline/medium_facenet_tutorial/train_classifier.py \
   --input-dir /facial_recognition_pipeline/output/intermediate \
   --model-path /facial_recognition_pipeline/etc/20170511-185253/20170511-185253.pb \
   --classifier-path /facial_recognition_pipeline/output/classifier.pkl \
   --num-threads 16 \
   --num-epochs 25 \
   --min-num-images-per-class 10 \
   --is-train

   Note: Adjust --num-threads and --num-epochs based on your system's capabilities.

Evaluation

1. Evaluate the Classifier:

   After training, evaluate the classifier's performance.

References

• Building a Facial Recognition Pipeline with Deep Learning in TensorFlow
• TensorFlow Documentation
• Dlib Documentation