text_level is always just 1 in magic_pdf but not in deployed version on huggingface

[SaidKhudoyan]

Description of the bug | 错误描述

Hi, when running pdf extraction with magic-pdf the text_level is always just 1. However, when testing the online version on Huggingface, the text_levels are correctly shown. Is there any difference between the deployed version and the open-source python-api or is that just an erro?

How to reproduce the bug | 如何复现

Just install magic-pdf and try to run below code on some example pdf document.

os.environ["MINERU_TOOLS_CONFIG_JSON"] = os.path.join(os.path.dirname(os.path.abspath(__file__)), "magic-pdf.json")
from magic_pdf.data.data_reader_writer import FileBasedDataWriter, FileBasedDataReader
from magic_pdf.data.dataset import PymuDocDataset
from magic_pdf.model.doc_analyze_by_custom_model import doc_analyze
from magic_pdf.config.enums import SupportedPdfParseMethod

def extract_pdf_content(pdf_file_path):
    output_dir = "Output"
    image_dir = os.path.join(output_dir, "images")
    os.makedirs(image_dir, exist_ok=True)
    image_writer = FileBasedDataWriter(image_dir)
    output_writer = FileBasedDataWriter(output_dir)
    reader = FileBasedDataReader("")
    pdf_bytes = reader.read(pdf_file_path)
    ds = PymuDocDataset(pdf_bytes)
    if ds.classify() == SupportedPdfParseMethod.OCR:
        infer_result = ds.apply(doc_analyze, ocr=True)
        pipe_result = infer_result.pipe_ocr_mode(image_writer)
    else:
        infer_result = ds.apply(doc_analyze, ocr=False)
        pipe_result = infer_result.pipe_txt_mode(image_writer)
    
    relative_image_dir = "images"  # Relativer Pfad für JSON-Referenzen
    pipe_result.dump_content_list(output_writer, "content_list.json", relative_image_dir)

if __name__ == "__main__":
    pdf_file = "ExamplePDF"  # Ihr PDF-Dateiname
    extract_pdf_content(pdf_file)

Operating system | 操作系统

Linux

Python version | Python 版本

3.10

Software version | 软件版本 (magic-pdf --version)

1.0.x

Device mode | 设备模式

cpu

[myhloli]

Title classification is currently a preview feature, available only on the online demo for testing accuracy and stability. It will be officially released in the stable version once the feature matures.