feat: add an experimental v2 parser to improve performance (#29)

---------

Signed-off-by: Peter Staar <[email protected]>
Signed-off-by: Michele Dolfi <[email protected]>
Signed-off-by: rmdg88 <[email protected]>
Signed-off-by: Christoph Auer <[email protected]>
Co-authored-by: Peter Staar <[email protected]>
Co-authored-by: Peter W. J. Staar <[email protected]>
Co-authored-by: Michele Dolfi <[email protected]>
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
Co-authored-by: Rui Dias Gomes <[email protected]>
Co-authored-by: Christoph Auer <[email protected]>
Co-authored-by: rmdg88 <[email protected]>
Co-authored-by: Michele Dolfi <[email protected]>

Author: 6 people

.github/scripts/build_rhel.sh

@@ -3,31 +3,49 @@
 set -e  # trigger failure on error - do not remove!
 set -x  # display command on output
 
-# Build the sdist
+# Build the Python package with Poetry
 poetry build -f sdist
 
-# Compile the wheel from sdist in centos stream
+USE_SYSTEM_DEPS="ON"
+
+docker build --progress=plain \
+             --build-arg USE_SYSTEM_DEPS="$USE_SYSTEM_DEPS" \
+             -f - . <<EOF
+    # syntax=docker/dockerfile:1
 
-docker build -f - . <<EOF
     FROM quay.io/centos/centos:stream9
+
     RUN dnf config-manager --set-enabled crb
-    # RUN dnf copr -y enable cheimes/deepsearch-glm rhel-9-x86_64
+    
+    RUN dnf copr -y enable cheimes/deepsearch-glm rhel-9-x86_64
+
     RUN dnf install -y https://dl.fedoraproject.org/pub/epel/epel-release-latest-9.noarch.rpm \
        && dnf clean all
+
     RUN dnf install -y --nodocs \
-            gcc gcc-c++ git make cmake pkgconfig glibc-devel \
+            autoconf automake binutils cmake gcc gcc-c++ git glibc-devel glibc-headers glibc-static kernel-devel libtool libstdc++-devel make ninja-build pkgconfig zlib-devel \
             python3.11 python3.11-pip python3.11-devel \
-            libjpeg-turbo-devel libpng-devel qpdf-devel json-devel utf8cpp-devel zlib-devel \
-            loguru-devel  \
+            libjpeg-turbo-devel libpng-devel qpdf-devel json-devel utf8cpp-devel \
         && dnf clean all
-    
+
+    # # RUN dnf install -y --nodocs loguru-devel
+
+    # TEMPORARY loguru install method
+    # https://koji.fedoraproject.org/koji/buildinfo?buildID=2563067
+    RUN curl -O https://kojipkgs.fedoraproject.org//packages/loguru/2.2.0%5E20230406git4adaa18/5.el9/x86_64/loguru-2.2.0%5E20230406git4adaa18-5.el9.x86_64.rpm
+    RUN dnf install -y loguru-2.2.0%5E20230406git4adaa18-5.el9.x86_64.rpm
+    RUN curl -O https://kojipkgs.fedoraproject.org//packages/loguru/2.2.0%5E20230406git4adaa18/5.el9/x86_64/loguru-devel-2.2.0%5E20230406git4adaa18-5.el9.x86_64.rpm
+    RUN dnf install -y loguru-devel-2.2.0%5E20230406git4adaa18-5.el9.x86_64.rpm
+
     RUN mkdir /src
+
     COPY ./dist/*.tar.gz /src/
 
-    RUN USE_SYSTEM_DEPS=ON pip3.11 install /src/docling_parse*.tar.gz \
-        && python3.11 -c 'from docling_parse.docling_parse import pdf_parser'
+    RUN USE_SYSTEM_DEPS=\$USE_SYSTEM_DEPS pip3.11 install /src/docling_parse*.tar.gz \
+        && python3.11 -c 'from docling_parse.docling_parse import pdf_parser, pdf_parser_v2'
     
     COPY ./tests /src/tests
+
     RUN cd /src \
         && pip3.11 install pytest \
         && pytest

CMakeLists.txt

@@ -100,16 +100,16 @@ if(NOT USE_SYSTEM_DEPS)
     endif()
 endif()
 
-
 # include dependencies
+include(cmake/extlib_cxxopts.cmake)
+include(cmake/extlib_loguru.cmake)
 include(cmake/extlib_json.cmake)
 include(cmake/extlib_utf8.git.cmake)
 include(cmake/extlib_jpeg.cmake)
 include(cmake/extlib_qpdf_v11.cmake)
 
 # aggregate the targets created by the dependencies
-set(DEPENDENCIES qpdf jpeg utf8 json)
-
+set(DEPENDENCIES qpdf jpeg utf8 json loguru cxxopts)
 
 # ************************
 # *** libraries        ***
@@ -136,21 +136,53 @@ include_directories(${SUBDIRS})
 # ***  Executables   ***
 # **********************
 
-add_executable(parse.exe "${TOPLEVEL_PREFIX_PATH}/app/parse.cpp")
-set_property(TARGET parse.exe PROPERTY CXX_STANDARD 20)
-add_dependencies(parse.exe ${DEPENDENCIES})
-target_include_directories(parse.exe INTERFACE ${DEPENDENCIES})
-target_link_libraries(parse.exe ${DEPENDENCIES} ${LIB_LINK})
+string(REPLACE ";" ";" SUBDIRS_SEMICOLON "${SUBDIRS}")
+message(STATUS "subdirs: ${SUBDIRS_SEMICOLON}")
+
+string(REPLACE ";" ";" DEPENDENCIES_SEMICOLON "${DEPENDENCIES}")
+message(STATUS "cmake dependencies: ${DEPENDENCIES_SEMICOLON}")
+
+string(REPLACE ";" ";" LIBLINK_SEMICOLON "${LIB_LINK}")
+message(STATUS "cmake lib-link: ${LIBLINK_SEMICOLON}")
+
+add_executable(parse_v1.exe "${TOPLEVEL_PREFIX_PATH}/app/parse_v1.cpp")
+add_executable(parse_v2.exe "${TOPLEVEL_PREFIX_PATH}/app/parse_v2.cpp")
+add_executable(parse_v2_fonts.exe "${TOPLEVEL_PREFIX_PATH}/app/parse_v2_fonts.cpp")
+
+set_property(TARGET parse_v1.exe PROPERTY CXX_STANDARD 20)
+set_property(TARGET parse_v2.exe PROPERTY CXX_STANDARD 20)
+set_property(TARGET parse_v2_fonts.exe PROPERTY CXX_STANDARD 20)
+
+add_dependencies(parse_v1.exe ${DEPENDENCIES})
+add_dependencies(parse_v2.exe ${DEPENDENCIES})
+add_dependencies(parse_v2_fonts.exe ${DEPENDENCIES})
+
+target_include_directories(parse_v1.exe INTERFACE ${DEPENDENCIES})
+target_include_directories(parse_v2.exe INTERFACE ${DEPENDENCIES})
+target_include_directories(parse_v2_fonts.exe INTERFACE ${DEPENDENCIES})
+
+target_link_libraries(parse_v1.exe ${DEPENDENCIES} ${LIB_LINK})
+target_link_libraries(parse_v2.exe ${DEPENDENCIES} ${LIB_LINK})
+target_link_libraries(parse_v2_fonts.exe ${DEPENDENCIES} ${LIB_LINK})
 
 # **********************
 # ***  Libraries     ***
 # **********************
 
-add_library(libparse STATIC "${TOPLEVEL_PREFIX_PATH}/app/parse.cpp")
-add_dependencies(libparse ${DEPENDENCIES})
-target_include_directories(libparse INTERFACE ${DEPENDENCIES})
-set_target_properties(libparse PROPERTIES POSITION_INDEPENDENT_CODE ON)
-target_link_libraries(libparse ${DEPENDENCIES} ${LIB_LINK})
+add_library(parse_v1 STATIC "${TOPLEVEL_PREFIX_PATH}/app/parse_v1.cpp")
+add_library(parse_v2 STATIC "${TOPLEVEL_PREFIX_PATH}/app/parse_v1.cpp")
+
+add_dependencies(parse_v1 ${DEPENDENCIES})
+add_dependencies(parse_v2 ${DEPENDENCIES})
+
+target_include_directories(parse_v1 INTERFACE ${DEPENDENCIES})
+target_include_directories(parse_v2 INTERFACE ${DEPENDENCIES})
+
+set_target_properties(parse_v1 PROPERTIES POSITION_INDEPENDENT_CODE ON)
+set_target_properties(parse_v2 PROPERTIES POSITION_INDEPENDENT_CODE ON)
+
+target_link_libraries(parse_v1 ${DEPENDENCIES} ${LIB_LINK})
+target_link_libraries(parse_v2 ${DEPENDENCIES} ${LIB_LINK})
 
 # ***************************
 # ***  Python-binding     ***
@@ -161,25 +193,17 @@ target_link_libraries(libparse ${DEPENDENCIES} ${LIB_LINK})
 find_package(pybind11 CONFIG REQUIRED)
 
 pybind11_add_module(docling_parse "${TOPLEVEL_PREFIX_PATH}/app/pybind_parse.cpp")
-add_dependencies(docling_parse libparse)
+
+add_dependencies(docling_parse parse_v1 parse_v2)
+
 target_include_directories(docling_parse INTERFACE ${DEPENDENCIES})
+
 target_compile_definitions(docling_parse PRIVATE VERSION_INFO=${CMAKE_PROJECT_VERSION})
-target_link_libraries(docling_parse PRIVATE libparse)
+
+target_link_libraries(docling_parse PRIVATE parse_v1 parse_v2)
 
 # *****************
 # ***  Install  ***
 # *****************
 
 install(TARGETS docling_parse DESTINATION "${TOPLEVEL_PREFIX_PATH}/docling_parse")
-
-# *****************
-# ***  Testing  ***
-# *****************
-
-#function(do_test target arg result)
-#  add_test(NAME Comp${arg} COMMAND ${target} ${arg})
-#  set_tests_properties(Comp${arg} PROPERTIES PASS_REGULAR_EXPRESSION ${result})
-# endfunction()
-
-# do a bunch of result based tests
-# do_test(Tutorial 4 "4 is 2")

README.md

@@ -7,9 +7,32 @@
 [![Platforms](https://img.shields.io/badge/platform-macos%20|%20linux%20|%20windows-blue)](https://github.com/DS4SD/docling-parse/)
 [![License MIT](https://img.shields.io/github/license/DS4SD/docling-parse)](https://opensource.org/licenses/MIT)
 
-Simple package to extract text with coordinates from programmatic PDFs.
-This package is part of the [Docling](https://github.com/DS4SD/docling) conversion.
-
+Simple package to extract text, paths and bitmap images with coordinates from programmatic PDFs.
+This package is used in the [Docling](https://github.com/DS4SD/docling) PDF conversion.
+
+<table>
+  <tr>
+    <th>Version</th>
+    <th>Original</th>
+    <th>Word-level</th>
+    <th>Snippet-level</th>
+    <th>Performance</th>
+  </tr>
+  <tr>
+    <th>V1</th>
+    <td rowspan="2"><img src="./docs/example_visualisations/2305.14962v1.pdf_page=0.png" alt="screenshot" width="100"/></td>
+    <td>Not Supported</td>
+    <td><img src="./docs/example_visualisations/2305.14962v1.pdf_page=0.v1.png" alt="v1 snippet" width="100"/></td>
+    <td>~0.250 page/sec</td>
+  </tr>
+  <tr>
+    <th>V2</th>
+    <!-- The "Original" column image spans from the previous row -->
+    <td><img src="./docs/example_visualisations/2305.14962v1.pdf_page=0.v2.original.png" alt="v1 word" width="100"/></td>
+    <td><img src="./docs/example_visualisations/2305.14962v1.pdf_page=0.v2.sanitized.png" alt="v2 snippet" width="100"/></td>
+    <td>~0.050 page/sec <br><br>[~5-10X faster than v1]</td>
+  </tr>
+</table>
 
 ## Quick start
 
@@ -19,13 +42,13 @@ Install the package from Pypi
 pip install docling-parse
 ```
 
-Convert a PDF
+Convert a PDF (look in the [visualise.py](docling_parse/visualise.py) for a more detailed information)
 
 ```python
-from docling_parse.docling_parse import pdf_parser
+from docling_parse.docling_parse import pdf_parser_v2
 
 # Do this only once to load fonts (avoid initialising it many times)
-parser = pdf_parser()
+parser = pdf_parser_v2()
 
 # parser.set_loglevel(1) # 1=error, 2=warning, 3=success, 4=info
 
@@ -64,39 +87,7 @@ for page in range(0, num_pages):
     # parsed page is the first one!				  
     json_page = json_doc["pages"][0] 
 
-    page_dimensions = [json_page["dimensions"]["width"], json_page["dimensions"]["height"]]
-
-    # find text cells
-    cells=[]
-    for cell_id,cell in enumerate(json_page["cells"]):
-    	cells.append([page,
-	              cell_id,
-		      cell["content"]["rnormalized"], # text
-	              cell["box"]["device"][0], # x0 (lower left x)
-		      cell["box"]["device"][1], # y0 (lower left y)
-		      cell["box"]["device"][2], # x1 (upper right x)
-		      cell["box"]["device"][3], # y1 (upper right y)	
-		      ])
-
-    # find bitmap images
-    images=[]
-    for image_id,image in enumerate(json_page["images"]):
-    	images.append([page,
-	               image_id,
-	               image["box"][0], # x0 (lower left x)
-		       image["box"][1], # y0 (lower left y)
-		       image["box"][2], # x1 (upper right x)
-		       image["box"][3], # y1 (upper right y)
-		       ])
-
-    # find paths
-    paths=[]
-    for path_id,path in enumerate(json_page["paths"]):
-    	paths.append([page,
-	              path_id,
-	              path["x-values"], # array of x values
-	              path["y-values"], # array of y values
-		      ])
+	# <Insert your own code>
 
 # Unload the (QPDF) document and buffers
 parser.unload_document(doc_key)
@@ -128,10 +119,38 @@ To build the parse, simply run the following command in the root folder,
 rm -rf build; cmake -B ./build; cd build; make
 ```
 
-You can run the parser from your build folder with
+You can run the parser from your build folder. Example from parse_v1,
+
+```sh
+% ./parse_v1.exe -h
+A program to process PDF files or configuration files
+Usage:
+  PDFProcessor [OPTION...]
+
+  -i, --input arg          Input PDF file
+  -c, --config arg         Config file
+      --create-config arg  Create config file
+  -o, --output arg         Output file
+  -l, --loglevel arg       loglevel [error;warning;success;info]
+  -h, --help               Print usage
+```
+
+Example from parse_v2,
 
 ```sh
-./parse.exe <input-file> <optional-logging:true>
+% ./parse_v2.exe -h
+program to process PDF files or configuration files
+Usage:
+  PDFProcessor [OPTION...]
+
+  -i, --input arg          Input PDF file
+  -c, --config arg         Config file
+      --create-config arg  Create config file
+  -p, --page arg           Pages to process (default: -1 for all) (default:
+                           -1)
+  -o, --output arg         Output file
+  -l, --loglevel arg       loglevel [error;warning;success;info]
+  -h, --help               Print usage
 ```
 
 If you dont have an input file, then a template input file will be printed on the terminal.
@@ -148,7 +167,7 @@ poetry build
 To test the package, run,
 
 ```
-poetry run pytest ./tests/test_parse.py
+poetry run pytest ./tests -v -s
 ```
 
 
@@ -162,13 +181,15 @@ Please read [Contributing to Docling Parse](https://github.com/DS4SD/docling-par
 If you use Docling in your projects, please consider citing the following:
 
 ```bib
-@software{Docling,
-author = {Deep Search Team},
-month = {7},
-title = {{Docling}},
-url = {https://github.com/DS4SD/docling},
-version = {main},
-year = {2024}
+@techreport{Docling,
+  author = {Deep Search Team},
+  month = {8},
+  title = {Docling Technical Report},
+  url = {https://arxiv.org/abs/2408.09869},
+  eprint = {2408.09869},
+  doi = {10.48550/arXiv.2408.09869},
+  version = {1.0.0},
+  year = {2024}
 }
 ```