How to disable auto imgae OCR in service mode of PP-StructureV3？

[icejean]

According to issue#15785，

感谢您的反馈，这是一个已知问题，目前PP-StructureV3的后处理逻辑是当图片和图片中的文本有重叠的时候，会去除图片而不是保留两者，在未来版本中我们会支持该功能。您提供的图片无法展示的原因在于图片中的公式被检测到了，导致只保留了公式B-B，而去除了图片

I want to disable auto image OCR， and get the image back to the client, then perform an image OCR, because the image need to be saved.
In another way to do it directly with Python API call, auto image OCR is disabled and the image is returned:

import time
from paddleocr import PPStructureV3

pdf_path = "/home/demo/dataset/testPDF.pdf"
t1 = time.time()
pipeline = PPStructureV3(
    use_doc_orientation_classify=False,
    use_doc_unwarping=False,
    use_textline_orientation=True,
)
t2 =time.time()
output = pipeline.predict(input=pdf_path)
t3= time.time()
print("Load time: ",round(t2-t1,2))
print("Process time: ",round(t3-t2,2))
print("Total time: ",round(t3-t1,2))
for res in output:
    res.print()                       # 打印控制台结果
    res.save_to_markdown("/home/demo/temp/output_md")  # 保存为 Markdown
    res.save_to_json("/home/demo/temp/output_json")    # 保存结构化 JSON

So how can I disable auto image OCR in the sevice mode?

paddlex --serve --pipeline  /home/demo/.paddlex/configs/paddlex_config.yaml --port 8126

Here's the PP-StructureV3 service configuration file:
paddlex_config.txt

Thanks for your help!

[icejean]

It's O.K. to use the exported configure of

import time
from paddleocr import PPStructureV3

pdf_path = "/home/demo/dataset/testPDF.pdf"
t1 = time.time()
pipeline = PPStructureV3(
    use_doc_orientation_classify=False,
    use_doc_unwarping=False,
    use_textline_orientation=True,
)
t2 =time.time()
# 导出默认配置文件
pipeline.export_paddlex_config_to_yaml("/home/demo/.paddlex/configs/paddlex_config_ocr.yaml")

Then add max_num_input_imgs to the end to unlock the defalt 10 pages limitation:

Serving:
  extra:
    max_num_input_imgs: null

Then start service with this configuration:

paddlex --serve --pipeline  /home/demo/.paddlex/configs/paddlex_config_ocr.yaml 

The image will be returned with the OCR reuslt of it together.