SVG Converter Backend with StarVector

This project provides a backend service for converting images to SVG format using the StarVector-1B model from Hugging Face. It is designed to run locally or on cloud platforms like Google Colab, with a React frontend for user interaction.

Features

• Image to SVG conversion using StarVector-1B
• Local backend setup with Python, PyTorch, and Transformers
• Support for running on Google Colab with GPU
• React frontend integration via API

Requirements

• Python 3.8+
• PyTorch
• Transformers
• Pillow
• Other dependencies as per requirements.txt

Setup Instructions

1. Clone the repository.
2. Create and activate a Python virtual environment.
3. Install dependencies using pip.
4. Prepare a 224x224 RGB PNG image for testing.
5. Run the backend locally or on Google Colab.

Usage

• Local: Run main.py to convert a test image to SVG.
• Colab: Use the provided notebook to run the backend with GPU support.
• React frontend: Connect to the backend API exposed via ngrok.

Common Issues

• Token authentication errors: Ensure you have a valid Hugging Face token with read permissions.
• Model access: Request access to gated models on Hugging Face.
• Tensor size mismatches: Always resize input images to 224x224.
• FlashAttention errors on Windows: Disable flash attention via environment variables.
• GitHub push blocked due to secrets: Remove tokens from code and use environment variables.

Environment Variables

• HF_TOKEN: Hugging Face access token.
• USE_FLASH_ATTN=0: Disable FlashAttention for compatibility.
• TRANSFORMERS_ATTN_IMPLEMENTATION=eager: Force CPU-compatible attention.