📸 Face Tracker

📝 Description

This repository contains a face-tracking application that captures facial images, trains a model using the VGG16 architecture, and tests face detection in real-time.

📦 Requirements

Ensure you have the following installed:

• Python 3.x
• Jupyter Notebook
• TensorFlow
• OpenCV
• Albumentations
• Labelme

You can install all required packages using:

pip install tensorflow opencv-python albumentations labelme

🔧 How to Use the Application

1️⃣ Download the Repository

Clone or download the repository to your local machine:

git clone <repository_link>
cd <repository_directory>

2️⃣ Prepare the Data Folders

Before starting, delete all the photos and labels in the data and aug_data folders to ensure a fresh start.

3️⃣ Capture Facial Images

Run the following command to start capturing facial images using your webcam:

jupyter notebook data.ipynb

• This will prompt the webcam to capture photos of your face.

4️⃣ Annotate the Images

Use labelme to annotate the images by following these steps:

Start the Annotation Tool

Run the command:

labelme

This will open the Labelme interface.

Load the Images

• Navigate to the data/images/ folder where your captured images are stored.
• Click File → Change Output Directory and select data/labels/.
• Click File → Save Automatically to ensure annotations are saved automatically.

Annotate Each Image

• Click Edit → Create Rectangle to start drawing bounding boxes.
• Manually draw a box around your face for each image.
• When prompted, enter the class name (e.g., "face") and click OK.

Navigating Through Images

• Press D on your keyboard to move to the next image.
• If an image does not contain a face, skip it without annotating.

Once completed, labelme will generate JSON files in data/labels/, containing the bounding box coordinates and class labels.

5️⃣ Train the Model

Start the model training process by running:

jupyter notebook model.ipynb

• Images and labels will be sorted into train, test, and val folders.
• Augmentation will be applied using Albumentations, with the augmented images stored in aug_data/{train,test,val}/images and their corresponding labels saved in aug_data/{train,test,val}/labels.
• After training, the model will be saved as facetracker_vgg16.h5 in your Desktop directory.

6️⃣ Test the Face Detection App

To test face detection using the trained model, run:

jupyter notebook model_test.ipynb

• This will allow you to validate the face-tracking functionality in real time.

  

---

📂 Folder Structure

├── aug_data/         # Augmented images generated using Albumentations  
│   ├── train/       # Augmented training data  
│   │   ├── images/   # Augmented training images  
│   │   ├── labels/   # Corresponding JSON labels  
│   ├── test/        # Augmented test data  
│   │   ├── images/   # Augmented test images  
│   │   ├── labels/   # Corresponding JSON labels  
│   ├── val/         # Augmented validation data  
│   │   ├── images/   # Augmented validation images  
│   │   ├── labels/   # Corresponding JSON labels  
├── data/            # Raw captured images and annotations
│   ├── images       # Original directory containing the images before the train-test-val split
│   ├── labels       # Original directory containing the labels before the train-test-val split
│   ├── train/       # Training data  
│   │   ├── images/   # Raw training images  
│   │   ├── labels/   # JSON annotations from Labelme  
│   ├── test/        # Test data  
│   │   ├── images/   # Raw test images  
│   │   ├── labels/   # JSON annotations from Labelme  
│   ├── val/         # Validation data  
│   │   ├── images/   # Raw validation images  
│   │   ├── labels/   # JSON annotations from Labelme  
├── logs/            # Stores callback logs during model training  
├── data.ipynb       # Captures facial images and generates labels  
├── model.ipynb      # Trains the VGG16-based face tracking model  
├── model_test.ipynb # Tests the trained model for face detection  
├── README.md        # Project documentation  
├── .DS_Store        # macOS system file (safe to ignore)  


---

## 💡 Contributing  
Feel free to contribute by submitting **issues** or **pull requests**! 😊