🤖 Image Captioning System (BLIP-Powered)

An advanced AI-powered platform that automatically generates contextual, human-like captions for images using state-of-the-art Salesforce BLIP Transformer models.

---

🌟 System Overview

This project provide a robust solution for automated image-to-text generation. It transitions from a legacy EfficientNet+BiLSTM architecture to a modern, high-accuracy Transformer-based pipeline. Key objectives include:

• Accuracy: Leveraging Vision-Language Pre-training (VLP) for human-like descriptions.
• Scalability: Utilizing Celery + Redis to handle heavy ML inference asynchronously.
• Accessibility: Providing a seamless API for developers to integrate captioning into any website via a simple JS snippet.
• Modern UI: A premium dark-themed React frontend with real-time status polling.

---

🏗️ System Architecture & Diagrams

1. High-Level Logic Flow (Activity Diagram)

2. Async Communication (Sequence Diagram)

3. Database Schema (ERD Diagram)

4. Class Structure

5. Interaction (Use Case Diagram)

---

🚀 How to Use

1. Backend Setup

1. Environment: Create a virtual environment and install dependencies.

   python -m venv venv
   venv\Scripts\activate
   pip install -r backend/requirements.txt

2. Database: Run migrations.

   python manage.py migrate

3. Redis: Ensure Redis is running (default: localhost:6379).
4. Worker: Start the Celery worker in a separate terminal.

   celery -A backendImageCaption worker -l INFO --pool=solo

5. Server: Run the Django development server.

   python manage.py runserver

2. Frontend Setup

1. Install:

   cd frontend
   npm install

2. Start:

   npm start

3. API Integration

To add captions to any website, include the JS snippet found in the Documentation page. Replace YOUR_API_KEY_HERE with a valid key generated from the platform.

---

🛠️ Tools & Technologies

Layer	Tool / Framework	Version
Frontend	React	19.x
Backend	Django	5.x
API	Django REST Framework	3.15.x
Model	Salesforce BLIP (Base)	-
Inference	PyTorch / Transformers	5.3.0
Workers	Celery / Redis	5.x
Styling	Vanilla CSS / Bootstrap	5.3

---

⚖️ License & Hard Constraints

License

This project is licensed under the GNU General Public License v3.0 (GPL-3.0).

⚠️ HARD CONSTRAINTS

• DAILY QUOTA: Strictly capped at 1,000 requests per day per API key.
• COMMERCIAL USE: Commercial use of the Salesforce BLIP model weights MUST comply with the Salesforce BLIP license terms (BSD 3-Clause).
• ATTRIBUTION: You MUST maintain all existing copyright notices and "Powered by Salesforce BLIP" markers in any derivative works.
• IMAGE LIMITS: Max file size is 10MB. Supported: PNG, JPG, JPEG, GIF.
• WARRANTY: This is a research project. The authors provide NO WARRANTY and assume NO LIABILITY for model output.