Skip to content

GitLab

Commit e7649e43 authored by myhloli's avatar myhloli
Browse files

refactor: modularize PDF processing and enhance async support for VLM backend

parent f7b37684
Hide whitespace changes
Inline Side-by-side
Showing with 292 additions and 142 deletions
mineru/backend/vlm/vlm_analyze.py
View file @ e7649e43
...@@ -75,15 +75,15 @@ async def aio_doc_analyze( ...@@ -75,15 +75,15 @@ async def aio_doc_analyze(
if predictor is None: if predictor is None:
predictor = ModelSingleton().get_model(backend, model_path, server_url) predictor = ModelSingleton().get_model(backend, model_path, server_url)
load_images_start = time.time() # load_images_start = time.time()
images_list, pdf_doc = load_images_from_pdf(pdf_bytes) images_list, pdf_doc = load_images_from_pdf(pdf_bytes)
images_base64_list = [image_dict["img_base64"] for image_dict in images_list] images_base64_list = [image_dict["img_base64"] for image_dict in images_list]
load_images_time = round(time.time() - load_images_start, 2) # load_images_time = round(time.time() - load_images_start, 2)
logger.info(f"load images cost: {load_images_time}, speed: {round(len(images_base64_list)/load_images_time, 3)} images/s") # logger.info(f"load images cost: {load_images_time}, speed: {round(len(images_base64_list)/load_images_time, 3)} images/s")
infer_start = time.time() # infer_start = time.time()
results = await predictor.aio_batch_predict(images=images_base64_list) results = await predictor.aio_batch_predict(images=images_base64_list)
infer_time = round(time.time() - infer_start, 2) # infer_time = round(time.time() - infer_start, 2)
logger.info(f"infer finished, cost: {infer_time}, speed: {round(len(results)/infer_time, 3)} page/s") # logger.info(f"infer finished, cost: {infer_time}, speed: {round(len(results)/infer_time, 3)} page/s")
middle_json = result_to_middle_json(results, images_list, pdf_doc, image_writer) middle_json = result_to_middle_json(results, images_list, pdf_doc, image_writer)
return middle_json return middle_json, results
mineru/cli/common.py
View file @ e7649e43
...@@ -14,6 +14,7 @@ from mineru.utils.enum_class import MakeMode ...@@ -14,6 +14,7 @@ from mineru.utils.enum_class import MakeMode
from mineru.utils.pdf_image_tools import images_bytes_to_pdf_bytes from mineru.utils.pdf_image_tools import images_bytes_to_pdf_bytes
from mineru.backend.vlm.vlm_middle_json_mkcontent import union_make as vlm_union_make from mineru.backend.vlm.vlm_middle_json_mkcontent import union_make as vlm_union_make
from mineru.backend.vlm.vlm_analyze import doc_analyze as vlm_doc_analyze from mineru.backend.vlm.vlm_analyze import doc_analyze as vlm_doc_analyze
from mineru.backend.vlm.vlm_analyze import aio_doc_analyze as aio_vlm_doc_analyze
pdf_suffixes = [".pdf"] pdf_suffixes = [".pdf"]
image_suffixes = [".png", ".jpeg", ".jpg"] image_suffixes = [".png", ".jpeg", ".jpg"]
...@@ -73,155 +74,304 @@ def convert_pdf_bytes_to_bytes_by_pypdfium2(pdf_bytes, start_page_id=0, end_page ...@@ -73,155 +74,304 @@ def convert_pdf_bytes_to_bytes_by_pypdfium2(pdf_bytes, start_page_id=0, end_page
return output_bytes return output_bytes
def do_parse( def _prepare_pdf_bytes(pdf_bytes_list, start_page_id, end_page_id):
output_dir, """准备处理PDF字节数据"""
pdf_file_names: list[str], result = []
pdf_bytes_list: list[bytes], for pdf_bytes in pdf_bytes_list:
p_lang_list: list[str], new_pdf_bytes = convert_pdf_bytes_to_bytes_by_pypdfium2(pdf_bytes, start_page_id, end_page_id)
backend="pipeline", result.append(new_pdf_bytes)
parse_method="auto", return result
p_formula_enable=True,
p_table_enable=True,
server_url=None, def _process_output(
f_draw_layout_bbox=True, pdf_info,
f_draw_span_bbox=True, pdf_bytes,
f_dump_md=True, pdf_file_name,
f_dump_middle_json=True, local_md_dir,
f_dump_model_output=True, local_image_dir,
f_dump_orig_pdf=True, md_writer,
f_dump_content_list=True, f_draw_layout_bbox,
f_make_md_mode=MakeMode.MM_MD, f_draw_span_bbox,
start_page_id=0, f_dump_orig_pdf,
end_page_id=None, f_dump_md,
f_dump_content_list,
f_dump_middle_json,
f_dump_model_output,
f_make_md_mode,
middle_json,
model_output=None,
is_pipeline=True
): ):
from mineru.backend.pipeline.pipeline_middle_json_mkcontent import union_make as pipeline_union_make
"""处理输出文件"""
if f_draw_layout_bbox:
draw_layout_bbox(pdf_info, pdf_bytes, local_md_dir, f"{pdf_file_name}_layout.pdf")
if f_draw_span_bbox:
draw_span_bbox(pdf_info, pdf_bytes, local_md_dir, f"{pdf_file_name}_span.pdf")
if f_dump_orig_pdf:
md_writer.write(
f"{pdf_file_name}_origin.pdf",
pdf_bytes,
)
image_dir = str(os.path.basename(local_image_dir))
if f_dump_md:
make_func = pipeline_union_make if is_pipeline else vlm_union_make
md_content_str = make_func(pdf_info, f_make_md_mode, image_dir)
md_writer.write_string(
f"{pdf_file_name}.md",
md_content_str,
)
if f_dump_content_list:
make_func = pipeline_union_make if is_pipeline else vlm_union_make
content_list = make_func(pdf_info, MakeMode.CONTENT_LIST, image_dir)
md_writer.write_string(
f"{pdf_file_name}_content_list.json",
json.dumps(content_list, ensure_ascii=False, indent=4),
)
if f_dump_middle_json:
md_writer.write_string(
f"{pdf_file_name}_middle.json",
json.dumps(middle_json, ensure_ascii=False, indent=4),
)
if f_dump_model_output:
if is_pipeline:
md_writer.write_string(
f"{pdf_file_name}_model.json",
json.dumps(model_output, ensure_ascii=False, indent=4),
)
else:
output_text = ("\n" + "-" * 50 + "\n").join(model_output)
md_writer.write_string(
f"{pdf_file_name}_model_output.txt",
output_text,
)
logger.info(f"local output dir is {local_md_dir}")
def _process_pipeline(
output_dir,
pdf_file_names,
pdf_bytes_list,
p_lang_list,
parse_method,
p_formula_enable,
p_table_enable,
f_draw_layout_bbox,
f_draw_span_bbox,
f_dump_md,
f_dump_middle_json,
f_dump_model_output,
f_dump_orig_pdf,
f_dump_content_list,
f_make_md_mode,
):
"""处理pipeline后端逻辑"""
from mineru.backend.pipeline.model_json_to_middle_json import result_to_middle_json as pipeline_result_to_middle_json
from mineru.backend.pipeline.pipeline_analyze import doc_analyze as pipeline_doc_analyze
infer_results, all_image_lists, all_pdf_docs, lang_list, ocr_enabled_list = (
pipeline_doc_analyze(
pdf_bytes_list, p_lang_list, parse_method=parse_method,
formula_enable=p_formula_enable, table_enable=p_table_enable
)
)
for idx, model_list in enumerate(infer_results):
model_json = copy.deepcopy(model_list)
pdf_file_name = pdf_file_names[idx]
local_image_dir, local_md_dir = prepare_env(output_dir, pdf_file_name, parse_method)
image_writer, md_writer = FileBasedDataWriter(local_image_dir), FileBasedDataWriter(local_md_dir)
images_list = all_image_lists[idx]
pdf_doc = all_pdf_docs[idx]
_lang = lang_list[idx]
_ocr_enable = ocr_enabled_list[idx]
middle_json = pipeline_result_to_middle_json(
model_list, images_list, pdf_doc, image_writer,
_lang, _ocr_enable, p_formula_enable
)
pdf_info = middle_json["pdf_info"]
pdf_bytes = pdf_bytes_list[idx]
_process_output(
pdf_info, pdf_bytes, pdf_file_name, local_md_dir, local_image_dir,
md_writer, f_draw_layout_bbox, f_draw_span_bbox, f_dump_orig_pdf,
f_dump_md, f_dump_content_list, f_dump_middle_json, f_dump_model_output,
f_make_md_mode, middle_json, model_json, is_pipeline=True
)
async def _async_process_vlm(
output_dir,
pdf_file_names,
pdf_bytes_list,
backend,
f_draw_layout_bbox,
f_draw_span_bbox,
f_dump_md,
f_dump_middle_json,
f_dump_model_output,
f_dump_orig_pdf,
f_dump_content_list,
f_make_md_mode,
server_url=None,
):
"""异步处理VLM后端逻辑"""
parse_method = "vlm"
f_draw_span_bbox = False
for idx, pdf_bytes in enumerate(pdf_bytes_list):
pdf_file_name = pdf_file_names[idx]
local_image_dir, local_md_dir = prepare_env(output_dir, pdf_file_name, parse_method)
image_writer, md_writer = FileBasedDataWriter(local_image_dir), FileBasedDataWriter(local_md_dir)
middle_json, infer_result = await aio_vlm_doc_analyze(
pdf_bytes, image_writer=image_writer, backend=backend, server_url=server_url
)
pdf_info = middle_json["pdf_info"]
_process_output(
pdf_info, pdf_bytes, pdf_file_name, local_md_dir, local_image_dir,
md_writer, f_draw_layout_bbox, f_draw_span_bbox, f_dump_orig_pdf,
f_dump_md, f_dump_content_list, f_dump_middle_json, f_dump_model_output,
f_make_md_mode, middle_json, infer_result, is_pipeline=False
)
def _process_vlm(
output_dir,
pdf_file_names,
pdf_bytes_list,
backend,
f_draw_layout_bbox,
f_draw_span_bbox,
f_dump_md,
f_dump_middle_json,
f_dump_model_output,
f_dump_orig_pdf,
f_dump_content_list,
f_make_md_mode,
server_url=None,
):
"""同步处理VLM后端逻辑"""
parse_method = "vlm"
f_draw_span_bbox = False
if backend == "pipeline": for idx, pdf_bytes in enumerate(pdf_bytes_list):
pdf_file_name = pdf_file_names[idx]
from mineru.backend.pipeline.pipeline_middle_json_mkcontent import union_make as pipeline_union_make local_image_dir, local_md_dir = prepare_env(output_dir, pdf_file_name, parse_method)
from mineru.backend.pipeline.model_json_to_middle_json import result_to_middle_json as pipeline_result_to_middle_json image_writer, md_writer = FileBasedDataWriter(local_image_dir), FileBasedDataWriter(local_md_dir)
from mineru.backend.pipeline.pipeline_analyze import doc_analyze as pipeline_doc_analyze
for idx, pdf_bytes in enumerate(pdf_bytes_list):
new_pdf_bytes = convert_pdf_bytes_to_bytes_by_pypdfium2(pdf_bytes, start_page_id, end_page_id)
pdf_bytes_list[idx] = new_pdf_bytes
infer_results, all_image_lists, all_pdf_docs, lang_list, ocr_enabled_list = pipeline_doc_analyze(pdf_bytes_list, p_lang_list, parse_method=parse_method, formula_enable=p_formula_enable,table_enable=p_table_enable)
for idx, model_list in enumerate(infer_results):
model_json = copy.deepcopy(model_list)
pdf_file_name = pdf_file_names[idx]
local_image_dir, local_md_dir = prepare_env(output_dir, pdf_file_name, parse_method)
image_writer, md_writer = FileBasedDataWriter(local_image_dir), FileBasedDataWriter(local_md_dir)
images_list = all_image_lists[idx]
pdf_doc = all_pdf_docs[idx]
_lang = lang_list[idx]
_ocr_enable = ocr_enabled_list[idx]
middle_json = pipeline_result_to_middle_json(model_list, images_list, pdf_doc, image_writer, _lang, _ocr_enable, p_formula_enable)
pdf_info = middle_json["pdf_info"]
pdf_bytes = pdf_bytes_list[idx]
if f_draw_layout_bbox:
draw_layout_bbox(pdf_info, pdf_bytes, local_md_dir, f"{pdf_file_name}_layout.pdf")
if f_draw_span_bbox:
draw_span_bbox(pdf_info, pdf_bytes, local_md_dir, f"{pdf_file_name}_span.pdf")
if f_dump_orig_pdf: middle_json, infer_result = vlm_doc_analyze(
md_writer.write( pdf_bytes, image_writer=image_writer, backend=backend, server_url=server_url
f"{pdf_file_name}_origin.pdf", )
pdf_bytes,
)
if f_dump_md: pdf_info = middle_json["pdf_info"]
image_dir = str(os.path.basename(local_image_dir))
md_content_str = pipeline_union_make(pdf_info, f_make_md_mode, image_dir)
md_writer.write_string(
f"{pdf_file_name}.md",
md_content_str,
)
if f_dump_content_list: _process_output(
image_dir = str(os.path.basename(local_image_dir)) pdf_info, pdf_bytes, pdf_file_name, local_md_dir, local_image_dir,
content_list = pipeline_union_make(pdf_info, MakeMode.CONTENT_LIST, image_dir) md_writer, f_draw_layout_bbox, f_draw_span_bbox, f_dump_orig_pdf,
md_writer.write_string( f_dump_md, f_dump_content_list, f_dump_middle_json, f_dump_model_output,
f"{pdf_file_name}_content_list.json", f_make_md_mode, middle_json, infer_result, is_pipeline=False
json.dumps(content_list, ensure_ascii=False, indent=4), )
)
if f_dump_middle_json:
md_writer.write_string(
f"{pdf_file_name}_middle.json",
json.dumps(middle_json, ensure_ascii=False, indent=4),
)
if f_dump_model_output: def do_parse(
md_writer.write_string( output_dir,
f"{pdf_file_name}_model.json", pdf_file_names: list[str],
json.dumps(model_json, ensure_ascii=False, indent=4), pdf_bytes_list: list[bytes],
) p_lang_list: list[str],
backend="pipeline",
parse_method="auto",
p_formula_enable=True,
p_table_enable=True,
server_url=None,
f_draw_layout_bbox=True,
f_draw_span_bbox=True,
f_dump_md=True,
f_dump_middle_json=True,
f_dump_model_output=True,
f_dump_orig_pdf=True,
f_dump_content_list=True,
f_make_md_mode=MakeMode.MM_MD,
start_page_id=0,
end_page_id=None,
):
# 预处理PDF字节数据
pdf_bytes_list = _prepare_pdf_bytes(pdf_bytes_list, start_page_id, end_page_id)
logger.info(f"local output dir is {local_md_dir}") if backend == "pipeline":
_process_pipeline(
output_dir, pdf_file_names, pdf_bytes_list, p_lang_list,
parse_method, p_formula_enable, p_table_enable,
f_draw_layout_bbox, f_draw_span_bbox, f_dump_md, f_dump_middle_json,
f_dump_model_output, f_dump_orig_pdf, f_dump_content_list, f_make_md_mode
)
else: else:
if backend.startswith("vlm-"): if backend.startswith("vlm-"):
backend = backend[4:] backend = backend[4:]
f_draw_span_bbox = False _process_vlm(
parse_method = "vlm" output_dir, pdf_file_names, pdf_bytes_list, backend,
for idx, pdf_bytes in enumerate(pdf_bytes_list): f_draw_layout_bbox, f_draw_span_bbox, f_dump_md, f_dump_middle_json,
pdf_file_name = pdf_file_names[idx] f_dump_model_output, f_dump_orig_pdf, f_dump_content_list, f_make_md_mode,
pdf_bytes = convert_pdf_bytes_to_bytes_by_pypdfium2(pdf_bytes, start_page_id, end_page_id) server_url
local_image_dir, local_md_dir = prepare_env(output_dir, pdf_file_name, parse_method) )
image_writer, md_writer = FileBasedDataWriter(local_image_dir), FileBasedDataWriter(local_md_dir)
middle_json, infer_result = vlm_doc_analyze(pdf_bytes, image_writer=image_writer, backend=backend, server_url=server_url)
async def aio_do_parse(
pdf_info = middle_json["pdf_info"] output_dir,
pdf_file_names: list[str],
if f_draw_layout_bbox: pdf_bytes_list: list[bytes],
draw_layout_bbox(pdf_info, pdf_bytes, local_md_dir, f"{pdf_file_name}_layout.pdf") p_lang_list: list[str],
backend="pipeline",
if f_draw_span_bbox: parse_method="auto",
draw_span_bbox(pdf_info, pdf_bytes, local_md_dir, f"{pdf_file_name}_span.pdf") p_formula_enable=True,
p_table_enable=True,
if f_dump_orig_pdf: server_url=None,
md_writer.write( f_draw_layout_bbox=True,
f"{pdf_file_name}_origin.pdf", f_draw_span_bbox=True,
pdf_bytes, f_dump_md=True,
) f_dump_middle_json=True,
f_dump_model_output=True,
if f_dump_md: f_dump_orig_pdf=True,
image_dir = str(os.path.basename(local_image_dir)) f_dump_content_list=True,
md_content_str = vlm_union_make(pdf_info, f_make_md_mode, image_dir) f_make_md_mode=MakeMode.MM_MD,
md_writer.write_string( start_page_id=0,
f"{pdf_file_name}.md", end_page_id=None,
md_content_str, ):
) # 预处理PDF字节数据
pdf_bytes_list = _prepare_pdf_bytes(pdf_bytes_list, start_page_id, end_page_id)
if f_dump_content_list:
image_dir = str(os.path.basename(local_image_dir))
content_list = vlm_union_make(pdf_info, MakeMode.CONTENT_LIST, image_dir)
md_writer.write_string(
f"{pdf_file_name}_content_list.json",
json.dumps(content_list, ensure_ascii=False, indent=4),
)
if f_dump_middle_json:
md_writer.write_string(
f"{pdf_file_name}_middle.json",
json.dumps(middle_json, ensure_ascii=False, indent=4),
)
if f_dump_model_output: if backend == "pipeline":
model_output = ("\n" + "-" * 50 + "\n").join(infer_result) # pipeline模式暂不支持异步，使用同步处理方式
md_writer.write_string( _process_pipeline(
f"{pdf_file_name}_model_output.txt", output_dir, pdf_file_names, pdf_bytes_list, p_lang_list,
model_output, parse_method, p_formula_enable, p_table_enable,
) f_draw_layout_bbox, f_draw_span_bbox, f_dump_md, f_dump_middle_json,
f_dump_model_output, f_dump_orig_pdf, f_dump_content_list, f_make_md_mode
)
else:
if backend.startswith("vlm-"):
backend = backend[4:]
logger.info(f"local output dir is {local_md_dir}") await _async_process_vlm(
output_dir, pdf_file_names, pdf_bytes_list, backend,
f_draw_layout_bbox, f_draw_span_bbox, f_dump_md, f_dump_middle_json,
f_dump_model_output, f_dump_orig_pdf, f_dump_content_list, f_make_md_mode,
server_url
)
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment