GPU-Accelerated Text Recognition

A high-performance text extraction tool that leverages GPU acceleration and the Gemma 3:27B model through Ollama for efficient image-to-text conversion.

Features

• 🖼️ Image Processing: Supports PNG, JPG, JPEG, and WebP formats
• 🤖 AI-Powered: Utilizes Gemma 3:27B model for accurate text extraction
• 📊 Performance Metrics: Detailed GPU and processing statistics
• 🛠️ System Monitoring: GPU utilization, memory usage, and temperature tracking
• 🔍 Health Checks: Automatic service monitoring and error handling

Requirements

• NVIDIA GPU (for optimal performance) or compatible system
• Ollama framework installed
• Gemma 3:27B model available in Ollama
• Python 3.x
• Required Python packages:
  • ollama
  • Pillow (PIL)

Setup

1. Install Ollama and the Gemma 3:27B model:

   ollama pull gemma3:27b

2. Install Python dependencies:

   pip install ollama pillow

3. Create a samples directory and add your images

Usage

1. Place images in the samples directory
2. Run the script:

   python test.py

The script will:

• Process all images in the samples directory
• Sort images by size (smallest to largest)
• Extract text content
• Display detailed processing metrics
• Show GPU utilization statistics

Output

The script provides comprehensive output including:

• Image dimensions and file size
• Processing duration and token counts
• GPU utilization metrics
• Extracted text content
• Performance statistics

Notes

• Images are processed in order of size (smallest to largest)
• The model is configured to extract only text content, ignoring visual elements
• Processing time varies based on image size and complexity
• GPU monitoring is available for NVIDIA GPUs and macOS systems