AI Background Remover (End-to-End Full Stack MLOps Project)

🔗 Live on: https://removedbackground.duckdns.org This is the live production version of the project. You can directly upload your images and test the AI-based background removal system from this link.

TheProject.mp4

---

Project Description

This project is a full-stack, production-ready AI-powered background removal system. Users can authenticate with their Google accounts, upload images, remove image backgrounds using a state-of-the-art semantic segmentation deep learning model, and manage their processed image history through a secure cloud-based system.

The project is designed as a complete End-to-End MLOps application, covering:

• Model serving
• Secure authentication
• Database integration
• Frontend-backend communication
• Containerization
• Cloud deployment

---

Project Goal

The main goal of this project is to build a robust, scalable, and secure AI application that automatically removes image backgrounds using semantic segmentation. The focus is not only on machine learning performance but also on:

• Production-level system design
• Secure authentication
• Cloud infrastructure
• Real-world deployment practices
• Full MLOps lifecycle implementation

---

🛠️ Technologies Used

AI & Computer Vision

• Task: Semantic Segmentation (Foreground–Background Separation)
• Model Architecture: DeepLabV3+ (Latest generation segmentation architecture)
• Backbone: MobileNetV3-Large (Optimized for lightweight and CPU-based inference)
• Frameworks & Libraries: PyTorch, Torchvision, Albumentations, OpenCV

Backend (API & Business Logic)

• Framework: FastAPI
• Authentication: Google OAuth 2.0
• Authorization & Security: JWT (JSON Web Token)
• Session Handling: Secure HTTP-only Cookies
• Database: SQLite
• ORM: SQLAlchemy

Frontend (User Interface)

• Framework: Streamlit
• State & Cookie Management: extra-streamlit-components

DevOps, Container & Cloud

• Containerization: Docker, Docker Compose
• Cloud Provider: AWS EC2 (Ubuntu)
• Dynamic DNS: DuckDNS
• Network Routing: iptables Port Forwarding
• Authentication Flow: Google Login + Token-Based Access Control

Libraries Used

📂 Project Structure

• ai/

  • main.py → FastAPI application entry point
  • Deep learning model loader
  • Image preprocessing & inference pipeline

• backend/

  • auth.py → Google OAuth, JWT generation & validation
  • picture_operations.py → Image upload, processing & history endpoints

• frontend/

  • login_page.py → Google authentication interface
  • history_page.py → Displays past user operations
  • remove_background_page.py → Image upload & background removal interface
  • history_detail_page.py → Shows details of a selected generation

• db/

  • database.py → Database connection
  • tables.py → User & image metadata models

• Dockerfile

  • Optimized CPU-based PyTorch environment for production

• docker-compose.yml

  • Orchestration of Backend & Frontend containers

• .env

  • Environment variables (not pushed to GitHub)

---

How to Run Locally

You can run the system in two different ways:

• Using Docker (recommended)
• Running services manually without Docker

---

1. Environment Variable Configuration (.env)

Create a file named .env in the project root directory and configure it exactly like this:

# --- Google OAuth Configuration ---
GOOGLE_CLIENT_ID=your_google_client_id_here
GOOGLE_CLIENT_SECRET=your_google_client_secret_here

# --- Security ---
SECRET_KEY=your_random_secret_string
JWT_SECRET_KEY=your_random_jwt_secret_string

# --- Redirect Configuration ---
REDIRECT_URL=http://localhost:8000/auth/callback
FRONTEND_URL=http://localhost:8501

Without these variables:

• Google Login will NOT work
• Token security will break
• Backend authentication will fail

---

2. Option A: Run with Docker (Recommended)

This is the fastest and cleanest way to run the project.

Requirements

• Docker Desktop installed and running

Steps

docker-compose up --build

When you see:

Application startup complete

Open your browser:

http://localhost:8501

Backend will automatically run on:

http://localhost:8000

Everything (backend, frontend, AI model, database) runs inside containers.

---

3. Option B: Run Without Docker (Manual Setup)

You will need two separate terminals.

Install Dependencies

pip install -r requirements.txt

---

Terminal 1 – Start Backend

uvicorn backend.main:app --host 0.0.0.0 --port 8000 --reload

Backend runs at:

http://localhost:8000

---

Terminal 2 – Start Frontend

streamlit run frontend/remove_background_page.py

Frontend runs at:

http://localhost:8501

---

Authentication Flow

1. User logs in via Google OAuth
2. Backend generates a JWT token
3. Token is stored in secure HTTP-only cookies
4. All protected endpoints require valid JWT
5. User history is stored in SQLite

---

Production Deployment

• Deployed on AWS EC2 (Ubuntu)
• Served via DuckDNS
• Traffic routed through iptables
• Docker containers run backend and frontend as isolated services

---

Summary

This project is not just a background remover. It is a complete full-stack, cloud-deployed, authenticated, AI-powered MLOps system that demonstrates:

• Deep Learning in production
• Secure Authentication
• Containerization
• Cloud Networking
• Real-world deployment patterns