Cortex AI関数:AI_PARSE_DOCUMENTによるイメージ抽出¶

CREATE OR REPLACE PROCEDURE SAVE_EXTRACTED_IMAGES(r VARIANT)
RETURNS ARRAY
LANGUAGE PYTHON
RUNTIME_VERSION = '3.9'
PACKAGES = ('pillow', 'snowflake-snowpark-python')
HANDLER = 'run'
AS
$$
import base64
import io
import os
import tempfile
from PIL import Image

def process_parse_document_result(data: dict) -> tuple[str, str, str]:
    images = data["images"]
    for image in images:
        id = image["id"]
        data, image_base64 = image["image_base64"].split(";", 1)
        extension = data.split("/")[1]
        base64 = image_base64.split(",")[1]
        yield id, extension, base64

def decode_base64(encoded_image: str) -> bytes:
    return base64.b64decode(encoded_image)

def run(session, r):
    destination_path = r["DESTINATION_PATH"]
    parse_document_result = r["PARSE_DOCUMENT_RESULT"]

    if not destination_path:
        return ["Error: destination_path parameter is required"]
    if not destination_path.startswith("@"):
        return ["Error: destination_path must start with @ (e.g. @output_stage/path"]
    if destination_path == "@":
        return ["Error: destination_path must include a stage name after @"]

    # Clean the result directory
    session.sql(f"RM destination_path")

    uploaded_files = []
    with tempfile.TemporaryDirectory() as temp_dir:
        for image_id, extension, encoded_image in process_parse_document_result(parse_document_result):
            image_bytes = decode_base64(encoded_image)
            image: Image = Image.open(io.BytesIO(image_bytes))

            image_path = os.path.join(temp_dir, image_id)
            image.save(image_path)

            # Use session.file.put with source file path and auto_compress=False
            session.file.put(
                image_path, destination_path, auto_compress=False, overwrite=True
            )
            uploaded_files.append(f"{destination_path}/{image_id}")

            # Cleanup
            os.remove(image_path)
    return uploaded_files
$$;