AetherLens - Face Mesh (3D Landmark Mapping)

AetherLens - Face Mesh is a high-fidelity computer vision application that performs real-time 3D facial geometry tracking. By leveraging the MediaPipe Face Landmarker task, it maps 478 unique 3D landmarks onto the human face, providing a dense mesh suitable for facial analysis, augmented reality (AR) filters, and virtual avatars.

---

Key Features

• 478 3D Facial Landmarks: Detects a high-density mesh including eye contours, lips, and facial silhouettes.
• Expression Detection (Blendshapes): Capable of outputting 52 blendshape scores to recognize facial expressions like smiling, blinking, or brow movement.
• Real-Time Performance: Optimized for sub-millisecond processing on standard CPU/GPU hardware.
• Mirror Mode: Horizontally flipped feed for an intuitive user experience.

---

Installation & Setup

1. Clone the Repository

git clone https://github.com/NKumar-B/AetherLens-FaceMesh.git
cd AetherLens-FaceMesh

2. Set Up a Virtual Environment (Recommended)

python -m venv .venv
.\.venv\Scripts\activate

3. Install Dependencies

pip install -r requirements.txt

4. Download the Model Bundle

You must download the Face Landmarker model bundle from Google and place it in your project root:

• Model Name: face_landmarker.task
• Download Link: MediaPipe Face Landmarker Guide

---

How to Use

1. Run the application:

python FaceLandmarker.py

2. Face Mesh Visualization: Look into the camera to see the green landmark mesh track your face in real-time.
3. Exit: Press 'q' to close the window.

---

Technical Overview

The system uses a multi-stage ML pipeline:

1. Face Detection: A short-range BlazeFace model identifies the presence of a face.
2. Landmark Estimation: A 3D landmark model predicts 478 landmarks via regression on the detected face region.
3. Coordinate Mapping: Normalized coordinates are converted to pixel values based on the live camera resolution.

---

License

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

---

Acknowledgments

• Google MediaPipe: For providing the robust Face Landmarker Tasks API and pre-trained .task models.
• OpenCV (Open Source Computer Vision Library): For the powerful real-time image processing and visualization tools.
• The COCO Dataset Team: For their foundational work in standardizing computer vision training data.
• NumPy: For the efficient numerical processing required for coordinate mapping.

---