跳转至主要内容
Open In ColabOpen on GitHub

如何加载 PDF

可移植文档格式 (PDF),标准化为 ISO 32000,是由 Adobe 于 1992 年开发的文件格式,用于以独立于应用程序软件、硬件和操作系统的方式呈现文档,包括文本格式和图像。

本指南介绍了如何将 PDF 文档加载到 LangChain Document 格式中,供下游使用。

PDF 中的文本通常通过文本框表示。它们也可能包含图像。PDF 解析器可能会执行以下操作的某种组合

  • 通过启发式方法或 ML 推理将文本框聚合成行、段落和其他结构;
  • 对图像运行 OCR 以检测其中的文本;
  • 将文本分类为属于段落、列表、表格或其他结构;
  • 将文本结构化为表格行和列,或键值对。

LangChain 集成了许多 PDF 解析器。有些简单且相对底层;其他一些将支持 OCR 和图像处理,或执行高级文档布局分析。正确的选择将取决于您的需求。下面我们列举了可能性。

我们将在一个示例文件上演示这些方法

file_path = (
"../../docs/integrations/document_loaders/example_data/layout-parser-paper.pdf"
)
关于多模态模型的说明

许多现代 LLM 支持对多模态输入(例如,图像)进行推理。在某些应用中——例如对布局复杂、图表或扫描件的 PDF 进行问答——跳过 PDF 解析,而是将 PDF 页面转换为图像并直接传递给模型可能更有优势。我们在下面的多模态模型的使用部分演示了这方面的一个示例。

简单快速的文本提取

如果您正在寻找 PDF 中嵌入的文本的简单字符串表示形式,则以下方法是合适的。它将返回一个 Document 对象列表——每页一个——其中包含 Document 的 page_content 属性中页面文本的单个字符串。它不会解析图像或扫描 PDF 页面中的文本。在底层,它使用 pypdf Python 库。

LangChain 文档加载器 实现 lazy_load 及其异步变体 alazy_load,它们返回 Document 对象的迭代器。我们将在下面使用这些。

%pip install -qU pypdf
from langchain_community.document_loaders import PyPDFLoader

loader = PyPDFLoader(file_path)
pages = []
async for page in loader.alazy_load():
pages.append(page)
API 参考:PyPDFLoader
print(f"{pages[0].metadata}\n")
print(pages[0].page_content)
{'source': '../../docs/integrations/document_loaders/example_data/layout-parser-paper.pdf', 'page': 0}

LayoutParser : A Unified Toolkit for Deep
Learning Based Document Image Analysis
Zejiang Shen1( �), Ruochen Zhang2, Melissa Dell3, Benjamin Charles Germain
Lee4, Jacob Carlson3, and Weining Li5
1Allen Institute for AI
shannons@allenai.org
2Brown University
ruochen zhang@brown.edu
3Harvard University
{melissadell,jacob carlson }@fas.harvard.edu
4University of Washington
bcgl@cs.washington.edu
5University of Waterloo
w422li@uwaterloo.ca
Abstract. Recent advances in document image analysis (DIA) have been
primarily driven by the application of neural networks. Ideally, research
outcomes could be easily deployed in production and extended for further
investigation. However, various factors like loosely organized codebases
and sophisticated model configurations complicate the easy reuse of im-
portant innovations by a wide audience. Though there have been on-going
efforts to improve reusability and simplify deep learning (DL) model
development in disciplines like natural language processing and computer
vision, none of them are optimized for challenges in the domain of DIA.
This represents a major gap in the existing toolkit, as DIA is central to
academic research across a wide range of disciplines in the social sciences
and humanities. This paper introduces LayoutParser , an open-source
library for streamlining the usage of DL in DIA research and applica-
tions. The core LayoutParser library comes with a set of simple and
intuitive interfaces for applying and customizing DL models for layout de-
tection, character recognition, and many other document processing tasks.
To promote extensibility, LayoutParser also incorporates a community
platform for sharing both pre-trained models and full document digiti-
zation pipelines. We demonstrate that LayoutParser is helpful for both
lightweight and large-scale digitization pipelines in real-word use cases.
The library is publicly available at https://layout-parser.github.io .
Keywords: Document Image Analysis ·Deep Learning ·Layout Analysis
·Character Recognition ·Open Source library ·Toolkit.
1 Introduction
Deep Learning(DL)-based approaches are the state-of-the-art for a wide range of
document image analysis (DIA) tasks including document image classification [ 11,arXiv:2103.15348v2 [cs.CV] 21 Jun 2021

请注意,每个文档的元数据都存储了相应的页码。

基于 PDF 的向量搜索

一旦我们将 PDF 加载到 LangChain Document 对象中,我们就可以像往常一样索引它们(例如,RAG 应用程序)。下面我们使用 OpenAI 嵌入,尽管任何 LangChain embeddings 模型都足够了。

%pip install -qU langchain-openai
import getpass
import os

if "OPENAI_API_KEY" not in os.environ:
os.environ["OPENAI_API_KEY"] = getpass.getpass("OpenAI API Key:")
from langchain_core.vectorstores import InMemoryVectorStore
from langchain_openai import OpenAIEmbeddings

vector_store = InMemoryVectorStore.from_documents(pages, OpenAIEmbeddings())
docs = vector_store.similarity_search("What is LayoutParser?", k=2)
for doc in docs:
print(f'Page {doc.metadata["page"]}: {doc.page_content[:300]}\n')
Page 13: 14 Z. Shen et al.
6 Conclusion
LayoutParser provides a comprehensive toolkit for deep learning-based document
image analysis. The off-the-shelf library is easy to install, and can be used to
build flexible and accurate pipelines for processing documents with complicated
structures. It also supports hi

Page 0: LayoutParser : A Unified Toolkit for Deep
Learning Based Document Image Analysis
Zejiang Shen1( �), Ruochen Zhang2, Melissa Dell3, Benjamin Charles Germain
Lee4, Jacob Carlson3, and Weining Li5
1Allen Institute for AI
shannons@allenai.org
2Brown University
ruochen zhang@brown.edu
3Harvard University

布局分析和从图像中提取文本

如果您需要更细粒度的文本分割(例如,分割成不同的段落、标题、表格或其他结构)或需要从图像中提取文本,则以下方法是合适的。它将返回一个 Document 对象列表,其中每个对象代表页面上的一个结构。Document 的元数据存储页码和与对象相关的其他信息(例如,对于表格对象,它可能存储表格行和列)。

在底层,它使用 langchain-unstructured 库。有关使用 LangChain Unstructured 的更多信息,请参阅集成文档

Unstructured 支持 PDF 解析的多个参数

  • strategy (例如, "fast""hi-res")
  • API 或本地处理。您需要 API 密钥才能使用 API。

hi-res 策略提供对文档布局分析和 OCR 的支持。我们在下面通过 API 演示它。有关本地运行时需要考虑的事项,请参阅下面的本地解析部分。

%pip install -qU langchain-unstructured
import getpass
import os

if "UNSTRUCTURED_API_KEY" not in os.environ:
os.environ["UNSTRUCTURED_API_KEY"] = getpass.getpass("Unstructured API Key:")
Unstructured API Key: ········

和以前一样,我们初始化一个加载器并延迟加载文档

from langchain_unstructured import UnstructuredLoader

loader = UnstructuredLoader(
file_path=file_path,
strategy="hi_res",
partition_via_api=True,
coordinates=True,
)
docs = []
for doc in loader.lazy_load():
docs.append(doc)
INFO: Preparing to split document for partition.
INFO: Starting page number set to 1
INFO: Allow failed set to 0
INFO: Concurrency level set to 5
INFO: Splitting pages 1 to 16 (16 total)
INFO: Determined optimal split size of 4 pages.
INFO: Partitioning 4 files with 4 page(s) each.
INFO: Partitioning set #1 (pages 1-4).
INFO: Partitioning set #2 (pages 5-8).
INFO: Partitioning set #3 (pages 9-12).
INFO: Partitioning set #4 (pages 13-16).
INFO: HTTP Request: POST https://api.unstructuredapp.io/general/v0/general "HTTP/1.1 200 OK"
INFO: HTTP Request: POST https://api.unstructuredapp.io/general/v0/general "HTTP/1.1 200 OK"
INFO: HTTP Request: POST https://api.unstructuredapp.io/general/v0/general "HTTP/1.1 200 OK"
INFO: HTTP Request: POST https://api.unstructuredapp.io/general/v0/general "HTTP/1.1 200 OK"
INFO: Successfully partitioned set #1, elements added to the final result.
INFO: Successfully partitioned set #2, elements added to the final result.
INFO: Successfully partitioned set #3, elements added to the final result.
INFO: Successfully partitioned set #4, elements added to the final result.

在这里,我们从 16 页文档中恢复了 171 个不同的结构

print(len(docs))
171

我们可以使用文档元数据从单个页面恢复内容

first_page_docs = [doc for doc in docs if doc.metadata.get("page_number") == 1]

for doc in first_page_docs:
print(doc.page_content)
LayoutParser: A Unified Toolkit for Deep Learning Based Document Image Analysis
1 2 0 2 n u J 1 2 ] V C . s c [ 2 v 8 4 3 5 1 . 3 0 1 2 : v i X r a
Zejiang Shen® (<), Ruochen Zhang?, Melissa Dell®, Benjamin Charles Germain Lee?, Jacob Carlson®, and Weining Li®
1 Allen Institute for AI shannons@allenai.org 2 Brown University ruochen zhang@brown.edu 3 Harvard University {melissadell,jacob carlson}@fas.harvard.edu 4 University of Washington bcgl@cs.washington.edu 5 University of Waterloo w422li@uwaterloo.ca
Abstract. Recent advances in document image analysis (DIA) have been primarily driven by the application of neural networks. Ideally, research outcomes could be easily deployed in production and extended for further investigation. However, various factors like loosely organized codebases and sophisticated model configurations complicate the easy reuse of im- portant innovations by a wide audience. Though there have been on-going efforts to improve reusability and simplify deep learning (DL) model development in disciplines like natural language processing and computer vision, none of them are optimized for challenges in the domain of DIA. This represents a major gap in the existing toolkit, as DIA is central to academic research across a wide range of disciplines in the social sciences and humanities. This paper introduces LayoutParser, an open-source library for streamlining the usage of DL in DIA research and applica- tions. The core LayoutParser library comes with a set of simple and intuitive interfaces for applying and customizing DL models for layout de- tection, character recognition, and many other document processing tasks. To promote extensibility, LayoutParser also incorporates a community platform for sharing both pre-trained models and full document digiti- zation pipelines. We demonstrate that LayoutParser is helpful for both lightweight and large-scale digitization pipelines in real-word use cases. The library is publicly available at https://layout-parser.github.io.
Keywords: Document Image Analysis · Deep Learning · Layout Analysis · Character Recognition · Open Source library · Toolkit.
1 Introduction
Deep Learning(DL)-based approaches are the state-of-the-art for a wide range of document image analysis (DIA) tasks including document image classification [11,

提取表格和其他结构

我们加载的每个 Document 代表一个结构,例如标题、段落或表格。

某些结构可能对于索引或问答任务特别重要。这些结构可能是

  1. 分类以便于识别;
  2. 解析为更结构化的表示形式。

下面,我们识别并提取一个表格

单击以展开用于渲染页面的代码

%pip install -qU matplotlib PyMuPDF pillow

import fitz
import matplotlib.patches as patches
import matplotlib.pyplot as plt
from PIL import Image


def plot_pdf_with_boxes(pdf_page, segments):
pix = pdf_page.get_pixmap()
pil_image = Image.frombytes("RGB", [pix.width, pix.height], pix.samples)

fig, ax = plt.subplots(1, figsize=(10, 10))
ax.imshow(pil_image)
categories = set()
category_to_color = {
"Title": "orchid",
"Image": "forestgreen",
"Table": "tomato",
}
for segment in segments:
points = segment["coordinates"]["points"]
layout_width = segment["coordinates"]["layout_width"]
layout_height = segment["coordinates"]["layout_height"]
scaled_points = [
(x * pix.width / layout_width, y * pix.height / layout_height)
for x, y in points
]
box_color = category_to_color.get(segment["category"], "deepskyblue")
categories.add(segment["category"])
rect = patches.Polygon(
scaled_points, linewidth=1, edgecolor=box_color, facecolor="none"
)
ax.add_patch(rect)

# Make legend
legend_handles = [patches.Patch(color="deepskyblue", label="Text")]
for category in ["Title", "Image", "Table"]:
if category in categories:
legend_handles.append(
patches.Patch(color=category_to_color[category], label=category)
)
ax.axis("off")
ax.legend(handles=legend_handles, loc="upper right")
plt.tight_layout()
plt.show()


def render_page(doc_list: list, page_number: int, print_text=True) -> None:
pdf_page = fitz.open(file_path).load_page(page_number - 1)
page_docs = [
doc for doc in doc_list if doc.metadata.get("page_number") == page_number
]
segments = [doc.metadata for doc in page_docs]
plot_pdf_with_boxes(pdf_page, segments)
if print_text:
for doc in page_docs:
print(f"{doc.page_content}\n")
render_page(docs, 5)

LayoutParser: A Unified Toolkit for DL-Based DIA

5

Table 1: Current layout detection models in the LayoutParser model zoo

Dataset Base Model1 Large Model Notes PubLayNet [38] PRImA [3] Newspaper [17] TableBank [18] HJDataset [31] F / M M F F F / M M - - F - Layouts of modern scientific documents Layouts of scanned modern magazines and scientific reports Layouts of scanned US newspapers from the 20th century Table region on modern scientific and business document Layouts of history Japanese documents

1 For each dataset, we train several models of different sizes for different needs (the trade-off between accuracy vs. computational cost). For “base model” and “large model”, we refer to using the ResNet 50 or ResNet 101 backbones [13], respectively. One can train models of different architectures, like Faster R-CNN [28] (F) and Mask R-CNN [12] (M). For example, an F in the Large Model column indicates it has a Faster R-CNN model trained using the ResNet 101 backbone. The platform is maintained and a number of additions will be made to the model zoo in coming months.

layout data structures, which are optimized for efficiency and versatility. 3) When necessary, users can employ existing or customized OCR models via the unified API provided in the OCR module. 4) LayoutParser comes with a set of utility functions for the visualization and storage of the layout data. 5) LayoutParser is also highly customizable, via its integration with functions for layout data annotation and model training. We now provide detailed descriptions for each component.

3.1 Layout Detection Models

In LayoutParser, a layout model takes a document image as an input and generates a list of rectangular boxes for the target content regions. Different from traditional methods, it relies on deep convolutional neural networks rather than manually curated rules to identify content regions. It is formulated as an object detection problem and state-of-the-art models like Faster R-CNN [28] and Mask R-CNN [12] are used. This yields prediction results of high accuracy and makes it possible to build a concise, generalized interface for layout detection. LayoutParser, built upon Detectron2 [35], provides a minimal API that can perform layout detection with only four lines of code in Python:

1 import layoutparser as lp 2 image = cv2 . imread ( " image_file " ) # load images 3 model = lp . De t e c tro n2 Lay outM odel ( " lp :// PubLayNet / f as t er _ r c nn _ R _ 50 _ F P N_ 3 x / config " ) 4 5 layout = model . detect ( image )

LayoutParser provides a wealth of pre-trained model weights using various datasets covering different languages, time periods, and document types. Due to domain shift [7], the prediction performance can notably drop when models are ap- plied to target samples that are significantly different from the training dataset. As document structures and layouts vary greatly in different domains, it is important to select models trained on a dataset similar to the test samples. A semantic syntax is used for initializing the model weights in LayoutParser, using both the dataset name and model name lp://<dataset-name>/<model-architecture-name>.

请注意,尽管表格文本在文档内容中被折叠成单个字符串,但元数据包含其行和列的表示形式

from IPython.display import HTML, display

segments = [
doc.metadata
for doc in docs
if doc.metadata.get("page_number") == 5 and doc.metadata.get("category") == "Table"
]

display(HTML(segments[0]["text_as_html"]))
able 1. LUllclll 1ayoul actCCLloll 1110AdCs 111 L1C LayoOulralsel 1110U4cl 200
数据集| 基础模型'|注释
PubLayNet [38]F/M现代科学文档的布局
PRImAM扫描的现代杂志和科学报告的布局
报纸F20 世纪扫描的美国报纸的布局
TableBank [18]F现代科学和商业文档上的表格区域
HJDatasetF/M历史日本文档的布局

从特定部分提取文本

结构可能具有父子关系——例如,段落可能属于带有标题的部分。如果某个部分特别重要(例如,用于索引),我们可以隔离相应的 Document 对象。

下面,我们提取与文档“结论”部分关联的所有文本

render_page(docs, 14, print_text=False)

conclusion_docs = []
parent_id = -1
for doc in docs:
if doc.metadata["category"] == "Title" and "Conclusion" in doc.page_content:
parent_id = doc.metadata["element_id"]
if doc.metadata.get("parent_id") == parent_id:
conclusion_docs.append(doc)

for doc in conclusion_docs:
print(doc.page_content)
LayoutParser provides a comprehensive toolkit for deep learning-based document image analysis. The off-the-shelf library is easy to install, and can be used to build flexible and accurate pipelines for processing documents with complicated structures. It also supports high-level customization and enables easy labeling and training of DL models on unique document image datasets. The LayoutParser community platform facilitates sharing DL models and DIA pipelines, inviting discussion and promoting code reproducibility and reusability. The LayoutParser team is committed to keeping the library updated continuously and bringing the most recent advances in DL-based DIA, such as multi-modal document modeling [37, 36, 9] (an upcoming priority), to a diverse audience of end-users.
Acknowledgements We thank the anonymous reviewers for their comments and suggestions. This project is supported in part by NSF Grant OIA-2033558 and funding from the Harvard Data Science Initiative and Harvard Catalyst. Zejiang Shen thanks Doug Downey for suggestions.

从图像中提取文本

OCR 在图像上运行,从而可以提取其中的文本

render_page(docs, 11)

LayoutParser: A Unified Toolkit for DL-Based DIA

focuses on precision, efficiency, and robustness. The target documents may have complicated structures, and may require training multiple layout detection models to achieve the optimal accuracy. Light-weight pipelines are built for relatively simple documents, with an emphasis on development ease, speed and flexibility. Ideally one only needs to use existing resources, and model training should be avoided. Through two exemplar projects, we show how practitioners in both academia and industry can easily build such pipelines using LayoutParser and extract high-quality structured document data for their downstream tasks. The source code for these projects will be publicly available in the LayoutParser community hub.

11

5.1 A Comprehensive Historical Document Digitization Pipeline

The digitization of historical documents can unlock valuable data that can shed light on many important social, economic, and historical questions. Yet due to scan noises, page wearing, and the prevalence of complicated layout structures, ob- taining a structured representation of historical document scans is often extremely complicated. In this example, LayoutParser was used to develop a comprehensive pipeline, shown in Figure 5, to gener- ate high-quality structured data from historical Japanese firm financial ta- bles with complicated layouts. The pipeline applies two layout models to identify different levels of document structures and two customized OCR engines for optimized character recog- nition accuracy.

‘Active Learning Layout Annotate Layout Dataset | +—— Annotation Toolkit A4 Deep Learning Layout Layout Detection Model Training & Inference, A Post-processing — Handy Data Structures & \ Lo orajport 7 ) Al Pls for Layout Data A4 Default and Customized Text Recognition 0CR Models ¥ Visualization & Export Layout Structure Visualization & Storage The Japanese Document Helpful LayoutParser Modules Digitization Pipeline

As shown in Figure 4 (a), the document contains columns of text written vertically 15, a common style in Japanese. Due to scanning noise and archaic printing technology, the columns can be skewed or have vari- able widths, and hence cannot be eas- ily identified via rule-based methods. Within each column, words are sepa- rated by white spaces of variable size, and the vertical positions of objects can be an indicator of their layout type.

Fig. 5: Illustration of how LayoutParser helps with the historical document digi- tization pipeline.

15 A document page consists of eight rows like this. For simplicity we skip the row segmentation discussion and refer readers to the source code when available.

请注意,右侧图形中的文本被提取并合并到 Document 的内容中。

本地解析

本地解析需要安装其他依赖项。

Poppler (PDF 分析)

Tesseract (OCR)

我们还需要安装 unstructured PDF 扩展

%pip install -qU "unstructured[pdf]"

然后我们可以以大致相同的方式使用 UnstructuredLoader,放弃 API 密钥和 partition_via_api 设置

loader_local = UnstructuredLoader(
file_path=file_path,
strategy="hi_res",
)
docs_local = []
for doc in loader_local.lazy_load():
docs_local.append(doc)
WARNING: This function will be deprecated in a future release and `unstructured` will simply use the DEFAULT_MODEL from `unstructured_inference.model.base` to set default model name
INFO: Reading PDF for file: /Users/chestercurme/repos/langchain/libs/community/tests/integration_tests/examples/layout-parser-paper.pdf ...
INFO: Detecting page elements ...
INFO: Detecting page elements ...
INFO: Detecting page elements ...
INFO: Detecting page elements ...
INFO: Detecting page elements ...
INFO: Detecting page elements ...
INFO: Detecting page elements ...
INFO: Detecting page elements ...
INFO: Detecting page elements ...
INFO: Detecting page elements ...
INFO: Detecting page elements ...
INFO: Detecting page elements ...
INFO: Detecting page elements ...
INFO: Detecting page elements ...
INFO: Detecting page elements ...
INFO: Detecting page elements ...
INFO: Processing entire page OCR with tesseract...
INFO: Processing entire page OCR with tesseract...
INFO: Processing entire page OCR with tesseract...
INFO: Processing entire page OCR with tesseract...
INFO: Processing entire page OCR with tesseract...
INFO: Processing entire page OCR with tesseract...
INFO: padding image by 20 for structure detection
INFO: Processing entire page OCR with tesseract...
INFO: Processing entire page OCR with tesseract...
INFO: Processing entire page OCR with tesseract...
INFO: Processing entire page OCR with tesseract...
INFO: padding image by 20 for structure detection
INFO: Processing entire page OCR with tesseract...
INFO: Processing entire page OCR with tesseract...
INFO: Processing entire page OCR with tesseract...
INFO: Processing entire page OCR with tesseract...
INFO: Processing entire page OCR with tesseract...
INFO: Processing entire page OCR with tesseract...
INFO: Processing entire page OCR with tesseract...
INFO: Processing entire page OCR with tesseract...

然后可以像处理从 API 获得的文档一样处理文档列表。

多模态模型的使用

许多现代 LLM 支持对多模态输入(例如,图像)进行推理。在某些应用中——例如对布局复杂、图表或扫描件的 PDF 进行问答——跳过 PDF 解析,而是将 PDF 页面转换为图像并直接传递给模型可能更有优势。这允许模型对页面上的二维内容进行推理,而不是“一维”字符串表示形式。

原则上,我们可以使用任何支持多模态输入的 LangChain 聊天模型。这些模型的列表记录在此处。下面我们使用 OpenAI 的 gpt-4o-mini

首先,我们定义一个简短的实用函数,将 PDF 页面转换为 base64 编码的图像

%pip install -qU PyMuPDF pillow langchain-openai
import base64
import io

import fitz
from PIL import Image


def pdf_page_to_base64(pdf_path: str, page_number: int):
pdf_document = fitz.open(pdf_path)
page = pdf_document.load_page(page_number - 1) # input is one-indexed
pix = page.get_pixmap()
img = Image.frombytes("RGB", [pix.width, pix.height], pix.samples)

buffer = io.BytesIO()
img.save(buffer, format="PNG")

return base64.b64encode(buffer.getvalue()).decode("utf-8")
from IPython.display import Image as IPImage
from IPython.display import display

base64_image = pdf_page_to_base64(file_path, 11)
display(IPImage(data=base64.b64decode(base64_image)))

然后,我们可以以通常的方式查询模型。下面我们向它询问一个与页面上的图表相关的问题。

from langchain_openai import ChatOpenAI

llm = ChatOpenAI(model="gpt-4o-mini")
API 参考:ChatOpenAI
from langchain_core.messages import HumanMessage

query = "What is the name of the first step in the pipeline?"

message = HumanMessage(
content=[
{"type": "text", "text": query},
{
"type": "image_url",
"image_url": {"url": f"data:image/jpeg;base64,{base64_image}"},
},
],
)
response = llm.invoke([message])
print(response.content)
API 参考:HumanMessage
INFO: HTTP Request: POST https://api.openai.com/v1/chat/completions "HTTP/1.1 200 OK"
``````output
The first step in the pipeline is "Annotate Layout Dataset."

其他 PDF 加载器

有关可用的 LangChain PDF 加载器的列表,请参阅此表


此页面是否有所帮助?