AltTextGenerator

AltTextGenerator is a Python script designed to automatically generate alt text for images using the BLIP2 model. It also renames the images based on the generated alt text, ensuring accessible and SEO-friendly filenames.

Features

• Automatically generates alt text for images using the BLIP2 model.
• Renames image files based on the generated alt text.
• Supports a variety of image formats, including .jpg, .jpeg, .png, .gif, .jfif, .svg, and .webp.
• Automatically detects the device for processing (CUDA, MPS, or CPU).
• User-friendly interface for selecting the folder containing images.

Requirements

• Python 3.12 or above
• Pillow (>=11.1.0,<12.0.0) for image handling
• Hugging Face Transformers (>=4.48.0,<5.0.0) library
• torch (>=2.5.1,<3.0.0) library
• cairosvg (>=2.5.2,<3.0.0) for SVG image support

Installation

1. Clone the repository or download the script:

   git clone https://github.com/fvarghese99/AltTextGenerator.git

   Navigate to the project directory:

   cd /path/to/AltTextGenerator

2. Create a virtual environment with custom name 'AltTextGeneratorEnv' (recommended):

   python -m venv AltTextGeneratorEnv

   Activate the virtual environment:
   For macOS/Linux
   

    source AltTextGeneratorEnv/bin/activate  

   For Windows
   

   AltTextGeneratorEnv\Scripts\activate     

3. Install the required packages:

   pip install -r requirements.txt

PyTorch Dependencies

To ensure you install the correct versions of torch, torchvision, and torchaudio that are compatible with your system (e.g. with CUDA support or CPU-only), please refer to the official PyTorch Get Started Locally page.

For example, if you are using pip and wish to install a CUDA-enabled version (replace cu126 with the CUDA version suited to your system), you might run:

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu126

Usage

1. Place the images in the default img folder or create your own folder.
2. Run the script:

   python -m src.main

3. When prompted, enter the path to the folder containing images or press Enter to use the default img folder:

   Enter the path to the folder with images (or press Enter for './data/img'): 

4. The script will:
   • Generate alt text for each image.
   • Rename the images based on the generated alt text.
5. View the renamed images in the specified folder.

Testing

To ensure the script is working correctly, follow these steps to run the tests:

1. Install the pytest package:

   pip install pytest

2. Run all tests:

   pytest tests/

3. Test details:

   • The tests include validation for:
     • The BLIP2 model functionality.
     • Utility functions for file and image processing.
     • End-to-end integration of image processing and renaming.
       

4. Check the output to conform that all tests have passed. Example:

   ============================= test session starts ==============================
   platform darwin -- Python 3.12.8, pytest-7.4.4, pluggy-1.5.0
   rootdir: /AltTextGenerator
   plugins: cov-6.0.0
   collected 24 items
   
   tests/test_blip2.py ......                                              [ 25%]
   tests/test_file_utils.py ......                                         [ 50%]
   tests/test_image_utils.py ......                                        [ 75%]
   tests/test_main.py ......                                               [100%]
   
   ============================== 24 passed in 5.06s ===============================

Notes

• Images are resized to 512x512 pixels for compatibility with the BLIP2 model.
• Ensure that the folder path provided contains supported image formats.

Example Output

Enter the path to the folder with images (or press Enter for 'img'): 
Loading BLIP2 model...
Using MPS (Apple Silicon).
------------------------------------------------------------------
Image Path: img/img-1.jpg
Generated Alt Text: A brick building with a sign on the side

Renamed image to: img/A brick building with a sign on the side.jpg
------------------------------------------------------------------

Supported Image Formats

• .jpg
• .jpeg
• .png
• .gif
• .jfif
• .svg
• .webp

Acknowledgements

• Hugging Face Transformers
• PyTorch
• BLIP2