Image Captioning with Gradio

This repo provides a simple GUI application for using the Qwen2-VL-7B-Captioner-Relaxed model for image captioning. It is

Features

• Upload an image and generate a caption.
• Select from predefined system prompts or enter a custom prompt.
• Automatically load and initialize the model on the first caption generation.

Setup

Prerequisites

• Python 3.9 or higher
• pip (Python package installer)
• GPU with CUDA support with at least of 16GB VRAM
• 16 GB of storage for the model weights

Installation

1. Clone the repository:

   git clone https://github.com/ertugrul-dmr/qwen2vl-captioner-gui.git
   

   cd qwen2vl-captioner-gui
   

2. Install the required packages:

   For Linux:

   python3 -m venv venv
   source venv/bin/activate
   pip3 install torch
   pip3 install -r requirements.txt
   

   For Windows:

   python -m venv venv
   .\venv\Scripts\activate
   pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
   pip install -r requirements.txt
   

Running the Application

1. Run the app.py script:

   python app.py
   

2. Open your web browser and go to http://localhost:7860 to access the Gradio interface.

Usage

1. Upload an image by clicking on the image input box.
2. Select a system prompt from the dropdown menu or enter a custom prompt or leave as default.
3. Click the "Generate Caption" button to generate a caption for the image.
4. The generated caption will be displayed in the "Generated Caption" textbox.

License

This project is licensed under the MIT License. See the LICENSE file for details.

Acknowledgements

• Gradio
• Hugging Face Transformers
• Qwen2-VL