prepare.py

#!/usr/bin/env python3
"""Verify OneSearch-MI images and prepare ordered inference inputs."""

import argparse
import csv
import json
import struct
from concurrent.futures import ThreadPoolExecutor
from pathlib import Path
from urllib.parse import quote, urlsplit


def relative_file(root, value):
    path = Path(value)
    if path.is_absolute() or ".." in path.parts or "\\" in value:
        raise ValueError(f"Expected a relative image path: {value}")
    resolved = (root / path).resolve()
    if not resolved.is_relative_to(root.resolve()) or not resolved.is_file():
        raise ValueError(f"Missing image or path outside the dataset: {value}")
    return resolved


def load_rows(root):
    rows = [json.loads(line) for line in (root / "test.jsonl").read_text(encoding="utf-8").splitlines() if line.strip()]
    if not rows or len({row["id"] for row in rows}) != len(rows):
        raise ValueError("Sample IDs must be unique and the dataset must be nonempty")
    for row in rows:
        if (not isinstance(row["id"], str) or not row["id"].isdigit()
                or str(int(row["id"])) != row["id"]):
            raise ValueError("Sample IDs must be canonical nonnegative integer strings")
        if any(not isinstance(row[key], str) or not row[key].strip() for key in ("question", "answer")):
            raise ValueError("Question and answer must be nonempty strings")
        images = row["images"]
        if not isinstance(images, list) or len(images) < 2:
            raise ValueError("Each MI sample must contain at least two ordered image paths")
        if type(row["n_images"]) is not int or row["n_images"] != len(images):
            raise ValueError("n_images must equal the length of images")
        for image in images:
            relative_file(root, image)
    return rows


def prepare(root, rows, base_url, workers, verify_only=False):
    media = [json.loads(line) for line in (root / "media.jsonl").read_text(encoding="utf-8").splitlines() if line.strip()]
    inventory = {row["path"]: row for row in media}
    if len(inventory) != len(media):
        raise ValueError("Media inventory paths must be unique")
    referenced = {path for row in rows for path in row["images"]}
    if referenced != set(inventory):
        raise ValueError("Media inventory must match the referenced image set")

    def check(relative):
        path = relative_file(root, relative)
        info = inventory[relative]
        if path.stat().st_size != info["bytes"]:
            raise ValueError(f"Image byte size mismatch: {relative}")
        with path.open("rb") as handle:
            header = handle.read(24)
        if (len(header) != 24 or header[:8] != b"\x89PNG\r\n\x1a\n"
                or header[12:16] != b"IHDR"
                or struct.unpack(">II", header[16:24]) != (info["width"], info["height"])):
            raise ValueError(f"PNG dimensions differ from media.jsonl: {relative}")

    with ThreadPoolExecutor(max_workers=workers) as pool:
        list(pool.map(check, sorted(referenced)))
    result = {"samples": len(rows), "images": len(referenced), "verified": True}
    if verify_only:
        return result

    runtime = root / "runtime"
    runtime.mkdir(parents=True, exist_ok=True)
    with (runtime / "mi.csv").open("w", newline="", encoding="utf-8") as handle:
        writer = csv.DictWriter(handle, fieldnames=["index", "question", "answer"])
        writer.writeheader()
        writer.writerows({"index": row["id"], "question": row["question"], "answer": row["answer"]} for row in rows)
    with (runtime / "mi_manifest.jsonl").open("w", encoding="utf-8") as handle:
        for row in rows:
            record = {
                "case_id": row["id"], "ok": True, "modality": "multi_image",
                "n_images": len(row["images"]),
                "image_urls": [base_url.rstrip("/") + "/" + quote(path, safe="/") for path in row["images"]],
            }
            handle.write(json.dumps(record) + "\n")
    result["manifest"] = "runtime/mi_manifest.jsonl"
    return result


def main():
    parser = argparse.ArgumentParser(description=__doc__)
    parser.add_argument("--data-root", type=Path, default=Path(__file__).resolve().parent)
    parser.add_argument("--media-base-url", help="HTTP(S) base URL serving this MI dataset directory")
    parser.add_argument("--workers", type=int, default=4)
    parser.add_argument("--verify-only", action="store_true")
    args = parser.parse_args()
    if args.workers < 1:
        parser.error("--workers must be positive")
    if not args.verify_only:
        parsed = urlsplit(args.media_base_url or "")
        if parsed.scheme not in ("http", "https") or not parsed.netloc or parsed.query or parsed.fragment or parsed.username:
            parser.error("--media-base-url must be an HTTP(S) directory URL without credentials, query or fragment")
    root = args.data_root.resolve()
    print(json.dumps(prepare(root, load_rows(root), args.media_base_url, args.workers, args.verify_only)))


if __name__ == "__main__":
    main()