AI-Powered Image Captioning Web App

A clean, deployable AI project that generates natural language captions from uploaded images using Salesforce BLIP.

This repository is designed to be recruiter-friendly: practical AI integration, clear architecture, and readable code.

Project Overview

This app demonstrates a complete inference workflow:

1. User uploads an image in the Streamlit UI.
2. Image is processed with Pillow.
3. BLIP processor converts image into tensors.
4. BLIP model generates token IDs for a caption.
5. Tokens are decoded into readable text.
6. Caption is shown and can be downloaded.

Demo Screenshot

Main UI and generated caption preview:

Tech Stack

• Python
• Streamlit
• PyTorch
• Hugging Face Transformers
• Salesforce BLIP (Salesforce/blip-image-captioning-base)
• Pillow

How It Works

High-Level Architecture (for Recruiters)

The project uses a simple modular architecture:

• app.py: Streamlit presentation layer and user interactions.
• model.py: AI model loading and caption generation functions.
• utils.py: Image/file helper utilities.

This separation keeps the UI code clean while making model logic easier to test and maintain.

Inference Flow

1. app.py loads BLIP once using @st.cache_resource.
2. User uploads an image (or chooses one from example_images/).
3. utils.py converts the image to RGB PIL format.
4. model.py runs BLIP inference on CPU/GPU.
5. The generated caption is displayed and offered as a .txt download.

Project Structure

AI-Captioning-App/
├── app.py
├── model.py
├── utils.py
├── requirements.txt
├── README.md
├── assets/
│   ├── README.md
│   ├── p1.png
│   └── p2.png
└── example_images/
│   └── README.md

How to Run Locally

1. Clone the repository:

git clone https://github.com/asmiverma/AI-Captioning-App.git
cd AI-Captioning-App

2. Create a virtual environment:

python -m venv .venv

3. Activate the environment:

# Windows PowerShell
.\.venv\Scripts\Activate.ps1

# macOS/Linux
source .venv/bin/activate

4. Install dependencies:

pip install -r requirements.txt

5. Run the app:

streamlit run app.py

Future Improvements

• Support beam search/temperature options in the UI for caption diversity.
• Add multilingual caption translation.
• Add lightweight evaluation metrics over a small benchmark dataset.
• Package with Docker for one-command deployment.