Unmasked - Deepfake Detection System

Unmask the Fake. Protect the Real

Comment

🚨 Unmasked – Deepfake Detection System

🔥 Inspiration

Deepfakes today are not just viral entertainment they're weapons. They impersonate women, scam families, manipulate elections, and ruin reputations.

A close friend of ours once found a video circulating online his face, his expressions, his voice, but it wasn’t him. The helplessness, panic, and fear we saw in him left a mark on us.

We realized something important:

People aren’t losing to technology. They’re losing because they don’t know how to fight back.

So we built Unmasked a system that gives ordinary users the power to verify truth, detect manipulation, and take action. Not just a detector but a complete deepfake safety ecosystem.

🔍 What It Does

✅ 1. AI Deepfake Detection (Xception + Frame Analysis)

Users upload a video → backend extracts frames → passes them through a fine-tuned XceptionNet TensorFlow model → aggregates predictions → delivers:

  • Real vs Fake classification
  • Confidence metrics
  • Frame anomaly mapping
  • Manipulation indicators
  • Auto-generated PDF report (via ReportLab)

✅ 2. Personalized User Dashboard

Powered by Flask APIs + React Query:

  • Daily deepfake awareness tips
  • Previous analyses
  • Notifications
  • Reports
  • Activity stats

A digital literacy hub built for long-term awareness.

✅ 3. Community Forum

Using our SQLite-backed forum system:

  • Post questions
  • Share incidents
  • Discuss scams
  • Comment, like, reply
  • Admin moderation

A shared space for victims, learners & digital safety enthusiasts.

✅ 4. Support Center & Expert Assistance

A user can:

  • Request expert verification
  • File cybercrime complaints
  • Track complaint status
  • Access helpline numbers
  • Subscribe for safety updates

This bridges the real-world action gap—helping users move from detection → justice.

✅ 5. Deepfake News & Awareness Blogs

  • Curated blogs (seeded with content)
  • NewsAPI-powered live deepfake news feed
  • Daily safety tips

Continuous education to stay ahead of threats.

✅ 6. Admin Console

Admins get:

  • User analytics
  • Platform-wide insights
  • Dataset reset utilities
  • Report + forum moderation
  • Dashboard metrics (uploads, detections, posts)

⚠️ Challenges We Faced

🧠 Model Pipelines

Handling frame extraction, Xception preprocessing, and making inference fast without losing accuracy was tough. We optimized:

  • Frame limits
  • Batch prediction
  • Caching
  • Preprocessing overhead

🗂 Building a 15+ Table SQLite Ecosystem

Ensuring all flows—history, support center, blogs, forum, reports, notifications—were consistent and fast required a carefully indexed schema.

🔗 Frontend–Backend Sync

With protected routes, token verification, admin gating, and React Query caching, we had to ensure no broken flows or unauthorized access.

📝 Reliable PDF Reporting

Generating clean, consistent PDF reports with model outputs, stats, and timestamps demanded careful formatting.

🏗 How We Built It

⚡ Frontend (React + Vite + TypeScript)

  • shadcn/ui + Radix UI
  • TailwindCSS
  • React Router v6
  • React Query (API caching)
  • AuthContext with secure token storage
  • AdminRoute + ProtectedRoute wrappers
  • Optimized SPA layout with responsive sidebar & header

🧠 Backend (Flask + TensorFlow + SQLite)

  • Xception-based deepfake model
  • OpenCV frame extraction
  • Numpy preprocessing
  • ReportLab PDF generation
  • REST APIs for all modules
  • Scoped token-based authentication
  • DB seeding (admin, users, blogs, tips, posts)

🗂 Database (SQLite)

15+ tables:

users, analyses, history, reports, blogs, daily_tips,
notifications, subscriptions, expert_requests, complaints,
forum_posts, forum_comments, admin_logs …

Optimized with foreign keys & indexed fields.

🧪 Testing

  • Selenium end-to-end tests for login, upload, news & forum flows
  • Vite preview testing for UI integrity
  • Postman for report & model API validation

🏆 Accomplishments We’re Proud Of

🌐 A full end-to-end deepfake safety platform

Detection → Awareness → Reporting → Community → Admin moderation.

🧠 Explainable AI outputs

Our model doesn’t just say fake; it explains why, increasing trust.

🧩 Seamless multi-module integration

Support, blogs, forum, dashboard, news, reports all run on one coherent architecture.

📄 Instant PDF Evidence Reports

Users can attach these PDFs to cybercrime complaints or expert reviews.

🎓 What We Learned

✨ Safety is more than accuracy

Users need support, clarity, guidance—not just probabilities.

🧩 Building trustworthy AI means building transparent ecosystems

Explainability, reporting, and human review matter.

📦 Full-stack systems require careful flow orchestration

Especially with authentication, admin gates, and real-time UI updates.

🚀 What’s Next

🎙️ Audio deepfake detection

Integrating voice manipulation detection models.

📱 Mobile App

Real-time detection through phone camera.

🌏 Regional deepfake incident mapping

Track deepfake attacks geographically.

🧠 Multimodal deepfake analysis

Video + Audio + Metadata synergy.

🤖 Safety AI Assistant

A chatbot to guide users through detection, reporting, and emotional support.

Built With

Share this project:

Updates