MinerU/mineru/backend/pipeline/model_json_to_middle_json.py

# Copyright (c) Opendatalab. All rights reserved.
from mineru.utils.block_pre_proc import prepare_block_bboxes
from mineru.utils.pipeline_magic_model import MagicModel
from mineru.version import __version__
from mineru.utils.hash_utils import str_md5


def page_model_info_to_page_info(page_model_info, image_dict, page, image_writer, page_index, lang=None, ocr=False):
    scale = image_dict["scale"]
    page_pil_img = image_dict["img_pil"]
    page_img_md5 = str_md5(image_dict["img_base64"])
    page_w, page_h = map(int, page.get_size())
    magic_model = MagicModel(page_model_info, scale)

    """从magic_model对象中获取后面会用到的区块信息"""
    img_groups = magic_model.get_imgs()
    table_groups = magic_model.get_tables()

    """对image和table的区块分组"""
    img_body_blocks, img_caption_blocks, img_footnote_blocks = process_groups(
        img_groups, 'image_body', 'image_caption_list', 'image_footnote_list'
    )

    table_body_blocks, table_caption_blocks, table_footnote_blocks = process_groups(
        table_groups, 'table_body', 'table_caption_list', 'table_footnote_list'
    )

    discarded_blocks = magic_model.get_discarded()
    text_blocks = magic_model.get_text_blocks()
    title_blocks = magic_model.get_title_blocks()
    inline_equations, interline_equations, interline_equation_blocks = magic_model.get_equations()

    """将所有区块的bbox整理到一起"""
    interline_equation_blocks = []
    if len(interline_equation_blocks) > 0:
        all_bboxes, all_discarded_blocks, footnote_blocks = prepare_block_bboxes(
            img_body_blocks, img_caption_blocks, img_footnote_blocks,
            table_body_blocks, table_caption_blocks, table_footnote_blocks,
            discarded_blocks,
            text_blocks,
            title_blocks,
            interline_equation_blocks,
            page_w,
            page_h,
        )
    else:
        all_bboxes, all_discarded_blocks, footnote_blocks = prepare_block_bboxes(
            img_body_blocks, img_caption_blocks, img_footnote_blocks,
            table_body_blocks, table_caption_blocks, table_footnote_blocks,
            discarded_blocks,
            text_blocks,
            title_blocks,
            interline_equations,
            page_w,
            page_h,
        )


def result_to_middle_json(model_list, images_list, pdf_doc, image_writer, lang=None, ocr=False):
    middle_json = {"pdf_info": [], "_backend":"vlm", "_version_name": __version__}
    for page_index, page_model_info in enumerate(model_list):
        page = pdf_doc[page_index]
        image_dict = images_list[page_index]
        page_info = page_model_info_to_page_info(
            page_model_info, image_dict, page, image_writer, page_index, lang=lang, ocr=ocr
        )
        middle_json["pdf_info"].append(page_info)
    return middle_json


def process_groups(groups, body_key, caption_key, footnote_key):
    body_blocks = []
    caption_blocks = []
    footnote_blocks = []
    for i, group in enumerate(groups):
        group[body_key]['group_id'] = i
        body_blocks.append(group[body_key])
        for caption_block in group[caption_key]:
            caption_block['group_id'] = i
            caption_blocks.append(caption_block)
        for footnote_block in group[footnote_key]:
            footnote_block['group_id'] = i
            footnote_blocks.append(footnote_block)
    return body_blocks, caption_blocks, footnote_blocks