mirror of
https://github.com/infiniflow/ragflow.git
synced 2025-12-19 20:16:49 +08:00
Feature/mineru improvements (#11938)
我已在下面的评论中用中文重复说明。 ### What problem does this PR solve? ## Summary This PR enhances the MinerU document parser with additional configuration options, giving users more control over PDF parsing behavior and improving support for multilingual documents. ## Changes ### Backend (`deepdoc/parser/mineru_parser.py`) - Added configurable parsing options: - **Parse Method**: `auto`, `txt`, or `ocr` — allows users to choose the extraction strategy - **Formula Recognition**: Toggle for enabling/disabling formula extraction (useful to disable for Cyrillic documents where it may cause issues) - **Table Recognition**: Toggle for enabling/disabling table extraction - Added language code mapping (`LANGUAGE_TO_MINERU_MAP`) to translate RAGFlow language settings to MinerU-compatible language codes for better OCR accuracy - Improved parser configuration handling to pass these options through the processing pipeline ### Frontend (`web/`) - Created new `MinerUOptionsFormField` component that conditionally renders when MinerU is selected as the layout recognition engine - Added UI controls for: - Parse method selection (dropdown) - Formula recognition toggle (switch) - Table recognition toggle (switch) - Added i18n translations for English and Chinese - Integrated the options into both the dataset creation dialog and dataset settings page ### Integration - Updated `rag/app/naive.py` to forward MinerU options to the parser - Updated task service to handle the new configuration parameters ## Why MinerU is a powerful document parser, but the default settings don't work well for all document types. This PR allows users to: 1. Choose the best parsing method for their documents 2. Disable formula recognition for Cyrillic/non-Latin scripts where it causes issues 3. Control table extraction based on document needs 4. Benefit from automatic language detection for better OCR results ## Testing - [x] Tested MinerU parsing with different parse methods - [x] Verified UI renders correctly when MinerU is selected/deselected - [x] Confirmed settings persist correctly in dataset configuration ### Type of change - [x] Bug Fix (non-breaking change which fixes an issue) - [x] New Feature (non-breaking change which adds functionality) - [ ] Documentation Update - [x] Refactoring - [ ] Performance Improvement - [ ] Other (please describe): --------- Co-authored-by: user210 <user210@rt> Co-authored-by: Kevin Hu <kevinhu.sh@gmail.com>
This commit is contained in:
concertdictate
@ -24,6 +24,7 @@ import tempfile
|
||||
import threading
|
||||
import time
|
||||
import zipfile
|
||||
from dataclasses import dataclass
|
||||
from io import BytesIO
|
||||
from os import PathLike
|
||||
from pathlib import Path
|
||||
@ -53,6 +54,87 @@ class MinerUContentType(StrEnum):
|
||||
DISCARDED = "discarded"
|
||||
|
||||
|
||||
# Mapping from language names to MinerU language codes
|
||||
LANGUAGE_TO_MINERU_MAP = {
|
||||
'English': 'en',
|
||||
'Chinese': 'ch',
|
||||
'Traditional Chinese': 'chinese_cht',
|
||||
'Russian': 'east_slavic',
|
||||
'Ukrainian': 'east_slavic',
|
||||
'Indonesian': 'latin',
|
||||
'Spanish': 'latin',
|
||||
'Vietnamese': 'latin',
|
||||
'Japanese': 'japan',
|
||||
'Korean': 'korean',
|
||||
'Portuguese BR': 'latin',
|
||||
'German': 'latin',
|
||||
'French': 'latin',
|
||||
'Italian': 'latin',
|
||||
'Tamil': 'ta',
|
||||
'Telugu': 'te',
|
||||
'Kannada': 'ka',
|
||||
'Thai': 'th',
|
||||
'Greek': 'el',
|
||||
'Hindi': 'devanagari',
|
||||
}
|
||||
|
||||
|
||||
class MinerUBackend(StrEnum):
|
||||
"""MinerU processing backend options."""
|
||||
|
||||
PIPELINE = "pipeline" # Traditional multimodel pipeline (default)
|
||||
VLM_TRANSFORMERS = "vlm-transformers" # Vision-language model using HuggingFace Transformers
|
||||
VLM_MLX_ENGINE = "vlm-mlx-engine" # Faster, requires Apple Silicon and macOS 13.5+
|
||||
VLM_VLLM_ENGINE = "vlm-vllm-engine" # Local vLLM engine, requires local GPU
|
||||
VLM_VLLM_ASYNC_ENGINE = "vlm-vllm-async-engine" # Asynchronous vLLM engine, new in MinerU API
|
||||
VLM_LMDEPLOY_ENGINE = "vlm-lmdeploy-engine" # LMDeploy engine
|
||||
VLM_HTTP_CLIENT = "vlm-http-client" # HTTP client for remote vLLM server (CPU only)
|
||||
|
||||
|
||||
class MinerULanguage(StrEnum):
|
||||
"""MinerU supported languages for OCR (pipeline backend only)."""
|
||||
|
||||
CH = "ch" # Chinese
|
||||
CH_SERVER = "ch_server" # Chinese (server)
|
||||
CH_LITE = "ch_lite" # Chinese (lite)
|
||||
EN = "en" # English
|
||||
KOREAN = "korean" # Korean
|
||||
JAPAN = "japan" # Japanese
|
||||
CHINESE_CHT = "chinese_cht" # Chinese Traditional
|
||||
TA = "ta" # Tamil
|
||||
TE = "te" # Telugu
|
||||
KA = "ka" # Kannada
|
||||
TH = "th" # Thai
|
||||
EL = "el" # Greek
|
||||
LATIN = "latin" # Latin
|
||||
ARABIC = "arabic" # Arabic
|
||||
EAST_SLAVIC = "east_slavic" # East Slavic
|
||||
CYRILLIC = "cyrillic" # Cyrillic
|
||||
DEVANAGARI = "devanagari" # Devanagari
|
||||
|
||||
|
||||
class MinerUParseMethod(StrEnum):
|
||||
"""MinerU PDF parsing methods (pipeline backend only)."""
|
||||
|
||||
AUTO = "auto" # Automatically determine the method based on the file type
|
||||
TXT = "txt" # Use text extraction method
|
||||
OCR = "ocr" # Use OCR method for image-based PDFs
|
||||
|
||||
|
||||
@dataclass
|
||||
class MinerUParseOptions:
|
||||
"""Options for MinerU PDF parsing."""
|
||||
|
||||
backend: MinerUBackend = MinerUBackend.PIPELINE
|
||||
lang: Optional[MinerULanguage] = None # language for OCR (pipeline backend only)
|
||||
method: MinerUParseMethod = MinerUParseMethod.AUTO
|
||||
server_url: Optional[str] = None
|
||||
delete_output: bool = True
|
||||
parse_method: str = "raw"
|
||||
formula_enable: bool = True
|
||||
table_enable: bool = True
|
||||
|
||||
|
||||
class MinerUParser(RAGFlowPdfParser):
|
||||
def __init__(self, mineru_path: str = "mineru", mineru_api: str = "", mineru_server_url: str = ""):
|
||||
self.mineru_path = Path(mineru_path)
|
||||
@ -96,7 +178,8 @@ class MinerUParser(RAGFlowPdfParser):
|
||||
with open(full_path, "wb") as f:
|
||||
f.write(zip_ref.read(filename))
|
||||
|
||||
def _is_http_endpoint_valid(self, url, timeout=5):
|
||||
@staticmethod
|
||||
def _is_http_endpoint_valid(url, timeout=5):
|
||||
try:
|
||||
response = requests.head(url, timeout=timeout, allow_redirects=True)
|
||||
return response.status_code in [200, 301, 302, 307, 308]
|
||||
@ -141,7 +224,8 @@ class MinerUParser(RAGFlowPdfParser):
|
||||
self.logger.warning(f"[MinerU] vlm-http-client server check failed: {e}")
|
||||
try:
|
||||
response = requests.get(server_url, timeout=5)
|
||||
self.logger.info(f"[MinerU] vlm-http-client server connection check: success with status {response.status_code}")
|
||||
self.logger.info(
|
||||
f"[MinerU] vlm-http-client server connection check: success with status {response.status_code}")
|
||||
self.using_api = False
|
||||
return True, reason
|
||||
except Exception as e:
|
||||
@ -185,14 +269,15 @@ class MinerUParser(RAGFlowPdfParser):
|
||||
return False, reason
|
||||
|
||||
def _run_mineru(
|
||||
self, input_path: Path, output_dir: Path, method: str = "auto", backend: str = "pipeline", lang: Optional[str] = None, server_url: Optional[str] = None, callback: Optional[Callable] = None
|
||||
self, input_path: Path, output_dir: Path, options: MinerUParseOptions, callback: Optional[Callable] = None
|
||||
):
|
||||
if self.using_api:
|
||||
self._run_mineru_api(input_path, output_dir, method, backend, lang, callback)
|
||||
self._run_mineru_api(input_path, output_dir, options, callback)
|
||||
else:
|
||||
self._run_mineru_executable(input_path, output_dir, method, backend, lang, server_url, callback)
|
||||
self._run_mineru_executable(input_path, output_dir, options, callback)
|
||||
|
||||
def _run_mineru_api(self, input_path: Path, output_dir: Path, method: str = "auto", backend: str = "pipeline", lang: Optional[str] = None, callback: Optional[Callable] = None):
|
||||
def _run_mineru_api(self, input_path: Path, output_dir: Path, options: MinerUParseOptions,
|
||||
callback: Optional[Callable] = None):
|
||||
output_zip_path = os.path.join(str(output_dir), "output.zip")
|
||||
|
||||
pdf_file_path = str(input_path)
|
||||
@ -201,18 +286,18 @@ class MinerUParser(RAGFlowPdfParser):
|
||||
raise RuntimeError(f"[MinerU] PDF file not exists: {pdf_file_path}")
|
||||
|
||||
pdf_file_name = Path(pdf_file_path).stem.strip()
|
||||
output_path = os.path.join(str(output_dir), pdf_file_name, method)
|
||||
output_path = os.path.join(str(output_dir), pdf_file_name, options.method)
|
||||
os.makedirs(output_path, exist_ok=True)
|
||||
|
||||
files = {"files": (pdf_file_name + ".pdf", open(pdf_file_path, "rb"), "application/pdf")}
|
||||
|
||||
data = {
|
||||
"output_dir": "./output",
|
||||
"lang_list": lang,
|
||||
"backend": backend,
|
||||
"parse_method": method,
|
||||
"formula_enable": True,
|
||||
"table_enable": True,
|
||||
"lang_list": options.lang,
|
||||
"backend": options.backend,
|
||||
"parse_method": options.method,
|
||||
"formula_enable": options.formula_enable,
|
||||
"table_enable": options.table_enable,
|
||||
"server_url": None,
|
||||
"return_md": True,
|
||||
"return_middle_json": True,
|
||||
@ -229,7 +314,8 @@ class MinerUParser(RAGFlowPdfParser):
|
||||
self.logger.info(f"[MinerU] invoke api: {self.mineru_api}/file_parse")
|
||||
if callback:
|
||||
callback(0.20, f"[MinerU] invoke api: {self.mineru_api}/file_parse")
|
||||
response = requests.post(url=f"{self.mineru_api}/file_parse", files=files, data=data, headers=headers, timeout=1800)
|
||||
response = requests.post(url=f"{self.mineru_api}/file_parse", files=files, data=data, headers=headers,
|
||||
timeout=1800)
|
||||
|
||||
response.raise_for_status()
|
||||
if response.headers.get("Content-Type") == "application/zip":
|
||||
@ -253,15 +339,15 @@ class MinerUParser(RAGFlowPdfParser):
|
||||
self.logger.info("[MinerU] Api completed successfully.")
|
||||
|
||||
def _run_mineru_executable(
|
||||
self, input_path: Path, output_dir: Path, method: str = "auto", backend: str = "pipeline", lang: Optional[str] = None, server_url: Optional[str] = None, callback: Optional[Callable] = None
|
||||
self, input_path: Path, output_dir: Path, options: MinerUParseOptions, callback: Optional[Callable] = None
|
||||
):
|
||||
cmd = [str(self.mineru_path), "-p", str(input_path), "-o", str(output_dir), "-m", method]
|
||||
if backend:
|
||||
cmd.extend(["-b", backend])
|
||||
if lang:
|
||||
cmd.extend(["-l", lang])
|
||||
if server_url and backend == "vlm-http-client":
|
||||
cmd.extend(["-u", server_url])
|
||||
cmd = [str(self.mineru_path), "-p", str(input_path), "-o", str(output_dir), "-m", options.method]
|
||||
if options.backend:
|
||||
cmd.extend(["-b", options.backend])
|
||||
if options.lang:
|
||||
cmd.extend(["-l", options.lang])
|
||||
if options.server_url and options.backend == "vlm-http-client":
|
||||
cmd.extend(["-u", options.server_url])
|
||||
|
||||
self.logger.info(f"[MinerU] Running command: {' '.join(cmd)}")
|
||||
|
||||
@ -313,7 +399,8 @@ class MinerUParser(RAGFlowPdfParser):
|
||||
try:
|
||||
with pdfplumber.open(fnm) if isinstance(fnm, (str, PathLike)) else pdfplumber.open(BytesIO(fnm)) as pdf:
|
||||
self.pdf = pdf
|
||||
self.page_images = [p.to_image(resolution=72 * zoomin, antialias=True).original for _, p in enumerate(self.pdf.pages[page_from:page_to])]
|
||||
self.page_images = [p.to_image(resolution=72 * zoomin, antialias=True).original for _, p in
|
||||
enumerate(self.pdf.pages[page_from:page_to])]
|
||||
except Exception as e:
|
||||
self.page_images = None
|
||||
self.total_page = 0
|
||||
@ -375,7 +462,8 @@ class MinerUParser(RAGFlowPdfParser):
|
||||
pos = poss[-1]
|
||||
last_page_idx = pos[0][-1]
|
||||
if not (0 <= last_page_idx < page_count):
|
||||
self.logger.warning(f"[MinerU] Last page index {last_page_idx} out of range for {page_count} pages; skipping crop.")
|
||||
self.logger.warning(
|
||||
f"[MinerU] Last page index {last_page_idx} out of range for {page_count} pages; skipping crop.")
|
||||
if need_position:
|
||||
return None, None
|
||||
return
|
||||
@ -401,10 +489,12 @@ class MinerUParser(RAGFlowPdfParser):
|
||||
if 0 <= pn - 1 < page_count:
|
||||
bottom += self.page_images[pn - 1].size[1]
|
||||
else:
|
||||
self.logger.warning(f"[MinerU] Page index {pn}-1 out of range for {page_count} pages during crop; skipping height accumulation.")
|
||||
self.logger.warning(
|
||||
f"[MinerU] Page index {pn}-1 out of range for {page_count} pages during crop; skipping height accumulation.")
|
||||
|
||||
if not (0 <= pns[0] < page_count):
|
||||
self.logger.warning(f"[MinerU] Base page index {pns[0]} out of range for {page_count} pages during crop; skipping this segment.")
|
||||
self.logger.warning(
|
||||
f"[MinerU] Base page index {pns[0]} out of range for {page_count} pages during crop; skipping this segment.")
|
||||
continue
|
||||
|
||||
img0 = self.page_images[pns[0]]
|
||||
@ -417,7 +507,8 @@ class MinerUParser(RAGFlowPdfParser):
|
||||
bottom -= img0.size[1]
|
||||
for pn in pns[1:]:
|
||||
if not (0 <= pn < page_count):
|
||||
self.logger.warning(f"[MinerU] Page index {pn} out of range for {page_count} pages during crop; skipping this page.")
|
||||
self.logger.warning(
|
||||
f"[MinerU] Page index {pn} out of range for {page_count} pages during crop; skipping this page.")
|
||||
continue
|
||||
page = self.page_images[pn]
|
||||
x0, y0, x1, y1 = int(left), 0, int(right), int(min(bottom, page.size[1]))
|
||||
@ -461,7 +552,8 @@ class MinerUParser(RAGFlowPdfParser):
|
||||
poss.append(([int(p) - 1 for p in pn.split("-")], left, right, top, bottom))
|
||||
return poss
|
||||
|
||||
def _read_output(self, output_dir: Path, file_stem: str, method: str = "auto", backend: str = "pipeline") -> list[dict[str, Any]]:
|
||||
def _read_output(self, output_dir: Path, file_stem: str, method: str = "auto", backend: str = "pipeline") -> list[
|
||||
dict[str, Any]]:
|
||||
candidates = []
|
||||
seen = set()
|
||||
|
||||
@ -543,11 +635,13 @@ class MinerUParser(RAGFlowPdfParser):
|
||||
case MinerUContentType.TEXT:
|
||||
section = output["text"]
|
||||
case MinerUContentType.TABLE:
|
||||
section = output.get("table_body", "") + "\n".join(output.get("table_caption", [])) + "\n".join(output.get("table_footnote", []))
|
||||
section = output.get("table_body", "") + "\n".join(output.get("table_caption", [])) + "\n".join(
|
||||
output.get("table_footnote", []))
|
||||
if not section.strip():
|
||||
section = "FAILED TO PARSE TABLE"
|
||||
case MinerUContentType.IMAGE:
|
||||
section = "".join(output.get("image_caption", [])) + "\n" + "".join(output.get("image_footnote", []))
|
||||
section = "".join(output.get("image_caption", [])) + "\n" + "".join(
|
||||
output.get("image_footnote", []))
|
||||
case MinerUContentType.EQUATION:
|
||||
section = output["text"]
|
||||
case MinerUContentType.CODE:
|
||||
@ -569,24 +663,30 @@ class MinerUParser(RAGFlowPdfParser):
|
||||
return []
|
||||
|
||||
def parse_pdf(
|
||||
self,
|
||||
filepath: str | PathLike[str],
|
||||
binary: BytesIO | bytes,
|
||||
callback: Optional[Callable] = None,
|
||||
*,
|
||||
output_dir: Optional[str] = None,
|
||||
backend: str = "pipeline",
|
||||
lang: Optional[str] = None,
|
||||
method: str = "auto",
|
||||
server_url: Optional[str] = None,
|
||||
delete_output: bool = True,
|
||||
parse_method: str = "raw",
|
||||
self,
|
||||
filepath: str | PathLike[str],
|
||||
binary: BytesIO | bytes,
|
||||
callback: Optional[Callable] = None,
|
||||
*,
|
||||
output_dir: Optional[str] = None,
|
||||
backend: str = "pipeline",
|
||||
server_url: Optional[str] = None,
|
||||
delete_output: bool = True,
|
||||
parse_method: str = "raw",
|
||||
**kwargs,
|
||||
) -> tuple:
|
||||
import shutil
|
||||
|
||||
temp_pdf = None
|
||||
created_tmp_dir = False
|
||||
|
||||
# Assuming the dict is defined as shown
|
||||
lang = kwargs.get('lang', 'English')
|
||||
mineru_lang_code = LANGUAGE_TO_MINERU_MAP.get(lang, 'ch') # Returns 'ch' if lang not found
|
||||
mineru_method_raw_str = kwargs.get('parser_config', {}).get('mineru_parse_method', 'auto')
|
||||
enable_formula = kwargs.get('parser_config', {}).get('mineru_formula_enable', True)
|
||||
enable_table = kwargs.get('parser_config', {}).get('mineru_enable', True)
|
||||
|
||||
# remove spaces, or mineru crash, and _read_output fail too
|
||||
file_path = Path(filepath)
|
||||
pdf_file_name = file_path.stem.replace(" ", "") + ".pdf"
|
||||
@ -625,8 +725,18 @@ class MinerUParser(RAGFlowPdfParser):
|
||||
self.__images__(pdf, zoomin=1)
|
||||
|
||||
try:
|
||||
self._run_mineru(pdf, out_dir, method=method, backend=backend, lang=lang, server_url=server_url, callback=callback)
|
||||
outputs = self._read_output(out_dir, pdf.stem, method=method, backend=backend)
|
||||
options = MinerUParseOptions(
|
||||
backend=MinerUBackend(backend),
|
||||
lang=MinerULanguage(mineru_lang_code),
|
||||
method=MinerUParseMethod(mineru_method_raw_str),
|
||||
server_url=server_url,
|
||||
delete_output=delete_output,
|
||||
parse_method=parse_method,
|
||||
formula_enable=enable_formula,
|
||||
table_enable=enable_table,
|
||||
)
|
||||
self._run_mineru(pdf, out_dir, options, callback=callback)
|
||||
outputs = self._read_output(out_dir, pdf.stem, method=mineru_method_raw_str, backend=backend)
|
||||
self.logger.info(f"[MinerU] Parsed {len(outputs)} blocks from PDF.")
|
||||
if callback:
|
||||
callback(0.75, f"[MinerU] Parsed {len(outputs)} blocks from PDF.")
|
||||
|
||||
Reference in New Issue
Block a user