scripts/arabic_glm_ocr_extract.py

from __future__ import annotations

import argparse
import re
import sys
from pathlib import Path

TAG_RE = re.compile(r"<[^>]+>")
DEFAULT_ARABIC_GLM_OCR_MODEL = "sherif1313/Arabic-GLM-OCR-v2"


def clean_model_text(text: str) -> str:
    text = TAG_RE.sub("\n", text)
    text = re.sub(r"```(?:html|markdown|text|json)?", "", text, flags=re.IGNORECASE)
    text = text.replace("```", "")
    lines = [line.strip() for line in text.splitlines() if line.strip()]
    return "\n".join(lines)


def main() -> None:
    if hasattr(sys.stdout, "reconfigure"):
        sys.stdout.reconfigure(encoding="utf-8", errors="replace")
    if hasattr(sys.stderr, "reconfigure"):
        sys.stderr.reconfigure(encoding="utf-8", errors="replace")

    parser = argparse.ArgumentParser(description="Extract Arabic text from page images with Arabic-GLM-OCR.")
    parser.add_argument("--image-dir", required=True, type=Path)
    parser.add_argument("--out", required=True, type=Path)
    parser.add_argument("--model", default=DEFAULT_ARABIC_GLM_OCR_MODEL)
    parser.add_argument("--max-new-tokens", type=int, default=2048)
    args = parser.parse_args()

    image_paths = sorted(args.image_dir.glob("*.png"))
    total = max(len(image_paths), 1)
    print(f"ARABIC_READER_PROGRESS 0 {total}", flush=True)

    import torch
    from PIL import Image
    from transformers import AutoModelForImageTextToText, AutoProcessor

    device = "cuda" if torch.cuda.is_available() else "cpu"
    dtype = torch.float16 if device == "cuda" else torch.float32
    processor = AutoProcessor.from_pretrained(args.model)
    model = AutoModelForImageTextToText.from_pretrained(
        args.model,
        torch_dtype=dtype,
        device_map="auto" if device == "cuda" else None,
    )
    if device == "cpu":
        model.to(device)

    prompt = (
        "Extract the Arabic text exactly as it appears on this scanned page. "
        "Preserve reading order. Do not summarize, translate, explain, or correct the text."
    )
    pieces: list[str] = []
    image_paths = sorted(args.image_dir.glob("*.png"))
    total = max(len(image_paths), 1)
    for index, image_path in enumerate(image_paths, start=1):
        image = Image.open(image_path).convert("RGB")
        messages = [
            {
                "role": "user",
                "content": [
                    {"type": "image", "image": image},
                    {"type": "text", "text": prompt},
                ],
            }
        ]
        text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
        inputs = processor(text=[text], images=[image], return_tensors="pt").to(model.device)
        with torch.no_grad():
            output = model.generate(**inputs, max_new_tokens=args.max_new_tokens, do_sample=False)
        result = processor.decode(output[0][inputs["input_ids"].shape[1] :], skip_special_tokens=True)
        page_text = clean_model_text(result)
        if page_text:
            pieces.append(page_text)
        print(f"ARABIC_READER_PROGRESS {index} {total}", flush=True)

    args.out.parent.mkdir(parents=True, exist_ok=True)
    args.out.write_text("\n\n".join(pieces), encoding="utf-8")


if __name__ == "__main__":
    main()