Image to text (Llama 3.2 Vision & OCR)

This repo contains:

• image-text.py — send an image to a local or remote Ollama API (llama3.2-vision) and print streamed text.
• handwrite.py — TrOCR (microsoft/trocr-base-handwritten) for offline handwritten text (PyTorch + Hugging Face).
• image-to-text.ipynb — OpenCV + Tesseract experiments on sample images under image/.

Prerequisites

1. Python 3.10+ recommended.
2. Tesseract OCR installed and on your PATH (for the notebook and pytesseract). On Windows, install Tesseract and uncomment the tesseract_cmd line in the notebook if needed.
3. For image-text.py: Ollama running with llama3.2-vision (or compatible vision model), reachable at your API URL.

Setup

python -m venv .venv
.venv\Scripts\activate
pip install -r requirements.txt

Optional: copy .env.example to .env and set OLLAMA_URL, or export the variable in your shell.

Run

# From this directory; uses image/ and OLLAMA_URL (default in .env.example)
python image-text.py
python handwrite.py

Open image-to-text.ipynb in Jupyter or VS Code. Run the notebook from this project folder so paths like image/handnote-10.png resolve. The notebook is kept without stored outputs for cleaner Git diffs.

Git hygiene

Do not commit .env, API keys, or downloaded model weights. See .gitignore.