Hugging Face's logo

Spaces:
Hodfa71
/
vision-ai-engine
Sleeping

App Files Community
vision-ai-engine / README.md
hoda.fakhar
docs: add face verification screenshot to README
f7b4320
preview code
|
raw
history blame contribute delete
3.03 kB
metadata 
title: Vision AI Engine
emoji: 👁️‍🗨️
colorFrom: blue
colorTo: indigo
sdk: docker
pinned: false

📊 Biometric Case Study: Neural Occlusion Bypass

Match Proof Identity confirmed across hijab occlusion using deep face detection.

A professional-grade, Flask-powered facial biometric engine capable of real-time demographics analysis and cross-occlusion identity verification (e.g., matching across hijabs, glasses, and ages).

🔗 Live Deployment

▶ Launch Vision.AI Engine

Note: Hosted on Hugging Face Spaces (free tier). The first request may take 10–30 seconds as the app cold-starts and loads AI model weights into memory. Subsequent requests are fast.

🌟 Key Features

  • Neural Vision HUD (Live Mode): Real-time webcam analysis with biometric overlay — age, gender, emotion, and ethnicity.
  • Occlusion-Resistant Matcher: Matches identities across significant changes in headgear (hijab), eyewear, or lighting.
  • Deep Biometric Extraction: Gender, age range, ethnicity, and emotional state — powered by RetinaFace + DeepFace CNNs.
  • Neural Stabilization Buffer: Weighted voting across live frames to eliminate AI flicker.
  • Privacy-First: No images stored permanently; all analysis runs in volatile temp memory.

🛠️ Technology Stack

  • Engine: DeepFace — RetinaFace detector, VGG-Face/ArcFace models
  • Backend: Flask 3.0, TensorFlow 2.15.0, OpenCV (Headless)
  • Frontend: Vanilla HTML5/CSS3 (Glassmorphism), Font Awesome 6
  • Stability: Neural stabilization buffer with weighted voting across live frames
  • Deployment: Docker on Hugging Face Spaces, model weights pre-baked into image from HF Hub

🎨 Local Setup & Launch

  1. Initialize Environment:

    python -m venv venv
source venv/bin/activate  # or .\venv\Scripts\activate
pip install -r requirements.txt

  2. Download ML Weights:

    python scripts/download_weights.py

    Downloads all model weights into ~/.deepface/weights/. The app will not work without this step.

    Model Size Purpose
    age_model_weights.h5 ~514 MB Age estimation
    gender_model_weights.h5 ~514 MB Gender classification
    facial_expression_model_weights.h5 ~5 MB Emotion detection
    race_model_single_batch.h5 ~150 MB Ethnicity classification
    retinaface.h5 ~119 MB Face detection (accuracy-critical)

    RetinaFace is the face detector used by this app. It is significantly more accurate than basic OpenCV detection, especially for gender and age predictions.

  3. Run the App:

    python scripts/start_project.py

    Visit: http://localhost:5000/analysis/live

Created for a world-class Machine Learning Portfolio. Licensed under MIT.