KCAC Research Monorepo

This repository collects the local KCAC OCR and document-understanding projects into one GitHub repository.

The main project at the repository root is the KCAC OCR Pipeline v0.1. Additional related tools live under projects/.

Projects

Path	Purpose	Main documentation
.	OCR pipeline for formatting datasets, testing OCR/VLM engines, consensus, PAGE XML, eScriptorium export, reports, and Hugging Face JSONL export.	docs/how_to_run_and_use.md
projects/tesseract-ocr-training/	Tesseract, PaddleOCR, and OCR baseline training/evaluation scripts for Kurdish OCR experiments.	projects/tesseract-ocr-training/README.ne
projects/kcac-pdf-fetcher/	KCAC archive downloader/client for page images, PDFs, OCR text extraction, and validation.	projects/kcac-pdf-fetcher/README.ne
projects/surya-kurdish-region-detection/	Surya-based Kurdish text-line and layout-region detection workflow.	projects/surya-kurdish-region-detection/README.md

Large generated artifacts are intentionally not part of git: PDFs, page-image datasets, caches, trained weights, model checkpoints, output folders, and annotation outputs. Keep those locally or publish them separately as dataset/model releases.

OCR Pipeline Quick Start

Run from the repository root:

(& conda shell.powershell hook) | Out-String | Invoke-Expression
conda activate surya
python -m pip install -r requirements.txt
python -m pipeline --config config.yaml.example doctor

For your RTX 3090 and Ollama Qwen setup:

ollama list
ollama run qwen25vl-sorani-ocr:latest
python -m pipeline --config config.ollama.example doctor
python -m pipeline --config config.ollama.example bootstrap --limit 1

The full setup guide, file map, and reasoning are in docs/how_to_run_and_use.md.

OCR Pipeline Commands

python -m pipeline --config config.yaml bootstrap --limit 1
python -m pipeline --config config.yaml consensus --limit 1
python -m pipeline --config config.yaml pagexml --limit 1
python -m pipeline --config config.yaml escriptorium --limit 1
python -m pipeline --config config.yaml queue
python -m pipeline --config config.yaml reports
python -m pipeline --config config.yaml benchmark
python -m pipeline --config config.yaml hf-export

run-all runs the same stages in order. Existing per-engine/page outputs are skipped unless --force is passed.

Model And Data Notes

Kraken is the optional in-process legacy OCR dependency in requirements-ocr-py310.txt because it is not installable in Python 3.13. Calamari is split into requirements-calamari-py310.txt and should be installed in a separate environment because its python-bidi dependency conflicts with Kraken.

Use config.full.example only after you place real checkpoints under models/kraken/ and models/calamari/ or edit those paths. doctor is expected to fail for any enabled engine whose binary, checkpoint, Ollama server/model, or API key is missing.

KCAC Web Page

🚀 Open KCAC interactive page:
Click here to open home1.html

Source file: docs/home1.html