Datasets:

ArXiv:
License:
shinseung428 commited on
Commit
a53e9bc
1 Parent(s): b837da3

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +25 -8
README.md CHANGED
@@ -1,9 +1,10 @@
1
  # DP-Bench: Document Parsing Benchmark
2
 
3
  <div align="center">
4
- <img src="https://github.com/user-attachments/assets/54f8201b-5d3f-4673-b97d-848d8fcd10ff" width="800px">
5
  </div>
6
 
 
7
  Document parsing refers to the process of converting complex documents, such as PDFs and scanned images, into structured text formats like HTML and Markdown.
8
  It is especially useful as a preprocessor for [RAG](https://en.wikipedia.org/wiki/Retrieval-augmented_generation) systems, as it preserves key structural information from visually rich documents.
9
 
@@ -56,12 +57,16 @@ TEDS-S stands for Tree Edit Distance-based Similarity-Struct, measuring the stru
56
  ### Document sources
57
  The benchmark dataset is gathered from three sources: 90 samples from the Library of Congress; 90 samples from Open Educational Resources; and 20 samples from Upstage's internal resources, which is composed of company/product materials and [Solar 10.7B](https://arxiv.org/abs/2312.15166) paper. Together, these sources provide a broad and specialized range of information.
58
 
59
- | Sources | Count |
60
- |------------|-------|
 
 
61
  | Library of Congress | 90 |
62
  | Open educational resources | 90 |
63
  | Upstage | 20 |
64
 
 
 
65
  ### Layout elements
66
 
67
  While works like [ReadingBank](https://github.com/doc-analysis/ReadingBank) often focus solely on text conversion in document parsing, we have taken a more detailed approach by dividing the document into specific elements, with a particular emphasis on table performance.
@@ -72,8 +77,10 @@ Note that only Heading1 is included among various heading sizes because it repre
72
  This high-level segmentation is sufficient for assessing the core structure without adding unnecessary complexity.
73
  Detailed heading levels like Heading2 and Heading3 are omitted to keep the evaluation focused and manageable.
74
 
 
 
75
  | Category | Count |
76
- |------------|-------|
77
  | Paragraph | 804 |
78
  | Heading1 | 194 |
79
  | Footer | 168 |
@@ -87,6 +94,8 @@ Detailed heading levels like Heading2 and Heading3 are omitted to keep the evalu
87
  | Table | 55 |
88
  | Index | 10 |
89
 
 
 
90
  ### Dataset format
91
 
92
  The dataset is in JSON format, representing elements extracted from a PDF file, with each element defined by its position, layout class, and content.
@@ -130,9 +139,11 @@ Each element includes coordinates (x, y), a unique ID, and the page number it ap
130
  ...
131
  ```
132
 
 
 
133
  ### Document domains
134
  | Domain | Subdomain | Count |
135
- |-------------------------------------|-------------------------|-------|
136
  | Social Sciences | Economics | 26 |
137
  | | Political Science | 18 |
138
  | | Sociology | 16 |
@@ -149,6 +160,8 @@ Each element includes coordinates (x, y), a unique ID, and the page number it ap
149
  | | Computer Science | 8 |
150
  | | Statistics | 2 |
151
 
 
 
152
  ## Usage
153
 
154
  ### Setup
@@ -192,11 +205,15 @@ $ python evaluate.py \
192
  ```
193
 
194
  # Leaderboard
195
- | **Source** | **Date** | **TEDS** | **TEDS-S** | **NID** | **Avg Time (sec.)** |
196
- |----------------------|-----------|------------|-----------|---------------------|
 
 
197
  | aws | 24.09.26 | 90.22 | 95.94 | 14.47 |
198
  | llamaparse | 24.09.26 | 70.86 | 90.92 | 4.14 |
199
  | unstructured | 24.09.26 | 69.9 | 90.42 | 13.14 |
200
  | google | 24.09.26 | 68.75 | 90.09 | 5.85 |
201
  | microsoft | 24.09.26 | 89.07 | 87.03 | 4.44 |
202
- | upstage | 24.09.26 | **93.47** | **96.27** | **3.79** |
 
 
 
1
  # DP-Bench: Document Parsing Benchmark
2
 
3
  <div align="center">
4
+ <img src="https://cdn-uploads.huggingface.co/production/uploads/6524ab1e27d1f3d84ad07705/Q7CC2z4CAJzZ4-CGaSnBO.png" width="800px">
5
  </div>
6
 
7
+
8
  Document parsing refers to the process of converting complex documents, such as PDFs and scanned images, into structured text formats like HTML and Markdown.
9
  It is especially useful as a preprocessor for [RAG](https://en.wikipedia.org/wiki/Retrieval-augmented_generation) systems, as it preserves key structural information from visually rich documents.
10
 
 
57
  ### Document sources
58
  The benchmark dataset is gathered from three sources: 90 samples from the Library of Congress; 90 samples from Open Educational Resources; and 20 samples from Upstage's internal resources, which is composed of company/product materials and [Solar 10.7B](https://arxiv.org/abs/2312.15166) paper. Together, these sources provide a broad and specialized range of information.
59
 
60
+ <div style="width: 500px;">
61
+
62
+ | Sources | Count|
63
+ |:--------------------------:|:----:|
64
  | Library of Congress | 90 |
65
  | Open educational resources | 90 |
66
  | Upstage | 20 |
67
 
68
+ </div>
69
+
70
  ### Layout elements
71
 
72
  While works like [ReadingBank](https://github.com/doc-analysis/ReadingBank) often focus solely on text conversion in document parsing, we have taken a more detailed approach by dividing the document into specific elements, with a particular emphasis on table performance.
 
77
  This high-level segmentation is sufficient for assessing the core structure without adding unnecessary complexity.
78
  Detailed heading levels like Heading2 and Heading3 are omitted to keep the evaluation focused and manageable.
79
 
80
+ <div style="width: 500px;">
81
+
82
  | Category | Count |
83
+ |:----------:|:-----:|
84
  | Paragraph | 804 |
85
  | Heading1 | 194 |
86
  | Footer | 168 |
 
94
  | Table | 55 |
95
  | Index | 10 |
96
 
97
+ </div>
98
+
99
  ### Dataset format
100
 
101
  The dataset is in JSON format, representing elements extracted from a PDF file, with each element defined by its position, layout class, and content.
 
139
  ...
140
  ```
141
 
142
+ <div style="width: 800px;">
143
+
144
  ### Document domains
145
  | Domain | Subdomain | Count |
146
+ |:-----------------------------------:|:-----------------------:|:-----:|
147
  | Social Sciences | Economics | 26 |
148
  | | Political Science | 18 |
149
  | | Sociology | 16 |
 
160
  | | Computer Science | 8 |
161
  | | Statistics | 2 |
162
 
163
+ </div>
164
+
165
  ## Usage
166
 
167
  ### Setup
 
205
  ```
206
 
207
  # Leaderboard
208
+ <div style="width: 800px;">
209
+
210
+ | Source | Date | TEDS | TEDS-S | NID |
211
+ |:--------------------:|:---------:|:----------:|:---------:|:-------------------:|
212
  | aws | 24.09.26 | 90.22 | 95.94 | 14.47 |
213
  | llamaparse | 24.09.26 | 70.86 | 90.92 | 4.14 |
214
  | unstructured | 24.09.26 | 69.9 | 90.42 | 13.14 |
215
  | google | 24.09.26 | 68.75 | 90.09 | 5.85 |
216
  | microsoft | 24.09.26 | 89.07 | 87.03 | 4.44 |
217
+ | upstage | 24.09.26 | 93.47 | 96.27 | 3.79 |
218
+
219
+ </div>