Hugging Face's logo

Datasets:
upstage
/
dp-bench

ArXiv:
Tags:
nlp
Image-to-Text
License:
Dataset card Files Community
1
e9t commited on
Commit
1d7a9ba
•
1 Parent(s): 46d2f3a

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +38 -33
README.md CHANGED
@@ -1,4 +1,10 @@
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">
@@ -45,14 +51,13 @@ $$
45
  TEDS(T_a, T_b) = 1 - \frac{EditDist(T_a, T_b)}{\max(|T_a|, |T_b|)}
46
  $$
47
 
48
- The equation evaluates the similarity between two tables by modeling them as tree structures (T_a and T_b).
49
  This metric evaluates how accurately the table structure is predicted, including the content of each cell.
50
  A higher TEDS score indicates better overall performance in capturing both the table layout and the textual content.
51
 
52
  **TEDS-S (Tree Edit Distance-based Similarity-Struct).**
53
-
54
  TEDS-S stands for Tree Edit Distance-based Similarity-Struct, measuring the structural similarity between the predicted and reference tables.
55
- While the metric formulation is identical to TEDS, it uses modified tree representations, denoted as T_a' and T_b', where the nodes correspond solely to the table structure, omitting any cell-level content.
56
  This allows TEDS-S to concentrate on assessing the structural similarity of the tables, such as row and column alignment, without being influenced by the textual data contained within the cells.
57
  ## Benchmark dataset
58
 
@@ -62,7 +67,7 @@ The benchmark dataset is gathered from three sources: 90 samples from the Librar
62
  <div style="width: 500px;">
63
 
64
  | Sources | Count|
65
- |:--------------------------:|:----:|
66
  | Library of Congress | 90 |
67
  | Open educational resources | 90 |
68
  | Upstage | 20 |
@@ -82,7 +87,7 @@ Detailed heading levels like Heading2 and Heading3 are omitted to keep the evalu
82
  <div style="width: 500px;">
83
 
84
  | Category | Count |
85
- |:----------:|:-----:|
86
  | Paragraph | 804 |
87
  | Heading1 | 194 |
88
  | Footer | 168 |
@@ -144,23 +149,23 @@ Each element includes coordinates (x, y), a unique ID, and the page number it ap
144
  <div style="width: 800px;">
145
 
146
  ### Document domains
147
- | Domain | Subdomain | Count |
148
- |:-----------------------------------:|:-----------------------:|:-----:|
149
- | Social Sciences | Economics | 26 |
150
- | | Political Science | 18 |
151
- | | Sociology | 16 |
152
- | | Law | 12 |
153
- | | Cultural Anthropology | 11 |
154
- | | Education | 8 |
155
- | | Psychology | 4 |
156
- | Natural Sciences | Environmental Science | 26 |
157
- | | Biology | 10 |
158
- | | Astronomy | 4 |
159
- | Technology | Technology | 33 |
160
- | Mathematics and Information Sciences| Mathematics | 13 |
161
- | | Informatics | 9 |
162
- | | Computer Science | 8 |
163
- | | Statistics | 2 |
164
 
165
  </div>
166
 
@@ -171,8 +176,8 @@ Each element includes coordinates (x, y), a unique ID, and the page number it ap
171
  To set the environment, clone the repository and install the required dependencies by running the following commands:
172
 
173
  ```
174
- $ git clone httpshttps://huggingface.co/datasets/upstage/document-parse-benchmark.git
175
- $ cd document-parse-benchmark
176
  $ pip install -r requirements.txt
177
  ```
178
  The repository includes necessary scripts for performing inference and evaluation of document parsers.
@@ -209,13 +214,13 @@ $ python evaluate.py \
209
  # Leaderboard
210
  <div style="width: 800px;">
211
 
212
- | Source | Date | TEDS | TEDS-S | NID | Avg. Time |
213
- |:--------------------:|:---------:|:----------:|:---------:|:-----------:|:-----------:|
214
- | aws | 24.09.26 | 86.39 | 90.22 | 95.94 | 14.47 |
215
- | llamaparse | 24.09.26 | 68.9 | 70.86 | 90.92 | 4.14 |
216
- | unstructured | 24.09.26 | 64.49 | 69.9 | 90.42 | 13.14 |
217
- | google | 24.09.26 | 62.44 | 68.75 | 90.09 | 5.85 |
218
- | microsoft | 24.09.26 | 85.54 | 89.07 | 87.03 | 4.44 |
219
- | upstage | 24.09.26 | 91.01 | 93.47 | 96.27 | 3.79 |
220
 
221
  </div>
 
1
+ ---
2
+ license: mit
3
+ tags:
4
+ - nlp
5
+ ---
6
+
7
+ # **DP-Bench: Document Parsing Benchmark**
8
 
9
  <div align="center">
10
  <img src="https://cdn-uploads.huggingface.co/production/uploads/6524ab1e27d1f3d84ad07705/Q7CC2z4CAJzZ4-CGaSnBO.png" width="800px">
 
51
  TEDS(T_a, T_b) = 1 - \frac{EditDist(T_a, T_b)}{\max(|T_a|, |T_b|)}
52
  $$
53
 
54
+ The equation evaluates the similarity between two tables by modeling them as tree structures \\(T_a\\) and \\(T_b\\).
55
  This metric evaluates how accurately the table structure is predicted, including the content of each cell.
56
  A higher TEDS score indicates better overall performance in capturing both the table layout and the textual content.
57
 
58
  **TEDS-S (Tree Edit Distance-based Similarity-Struct).**
 
59
  TEDS-S stands for Tree Edit Distance-based Similarity-Struct, measuring the structural similarity between the predicted and reference tables.
60
+ While the metric formulation is identical to TEDS, it uses modified tree representations, denoted as \\(T_a'\\) and \\(T_b'\\), where the nodes correspond solely to the table structure, omitting any cell-level content.
61
  This allows TEDS-S to concentrate on assessing the structural similarity of the tables, such as row and column alignment, without being influenced by the textual data contained within the cells.
62
  ## Benchmark dataset
63
 
 
67
  <div style="width: 500px;">
68
 
69
  | Sources | Count|
70
+ |:---------------------------|:----:|
71
  | Library of Congress | 90 |
72
  | Open educational resources | 90 |
73
  | Upstage | 20 |
 
87
  <div style="width: 500px;">
88
 
89
  | Category | Count |
90
+ |:-----------|------:|
91
  | Paragraph | 804 |
92
  | Heading1 | 194 |
93
  | Footer | 168 |
 
149
  <div style="width: 800px;">
150
 
151
  ### Document domains
152
+ | Domain | Subdomain | Count |
153
+ |:-------------------------------------|:------------------------|------:|
154
+ | Social Sciences | Economics | 26 |
155
+ | Social Sciences | Political Science | 18 |
156
+ | Social Sciences | Sociology | 16 |
157
+ | Social Sciences | Law | 12 |
158
+ | Social Sciences | Cultural Anthropology | 11 |
159
+ | Social Sciences | Education | 8 |
160
+ | Social Sciences | Psychology | 4 |
161
+ | Natural Sciences | Environmental Science | 26 |
162
+ | Natural Sciences | Biology | 10 |
163
+ | Natural Sciences | Astronomy | 4 |
164
+ | Technology | Technology | 33 |
165
+ | Mathematics and Information Sciences | Mathematics | 13 |
166
+ | Mathematics and Information Sciences | Informatics | 9 |
167
+ | Mathematics and Information Sciences | Computer Science | 8 |
168
+ | Mathematics and Information Sciences | Statistics | 2 |
169
 
170
  </div>
171
 
 
176
  To set the environment, clone the repository and install the required dependencies by running the following commands:
177
 
178
  ```
179
+ $ git clone httpshttps://huggingface.co/datasets/upstage/dp-bench.git
180
+ $ cd dp-bench
181
  $ pip install -r requirements.txt
182
  ```
183
  The repository includes necessary scripts for performing inference and evaluation of document parsers.
 
214
  # Leaderboard
215
  <div style="width: 800px;">
216
 
217
+ | Source | Request date | TEDS | TEDS-S | NID | Avg. Time |
218
+ |:---------------------|:------------:|-----------:|----------:|------------:|------------:|
219
+ | aws | 2024-09-26 | 86.39 | 90.22 | 95.94 | 14.47 |
220
+ | llamaparse | 2024-09-26 | 68.90 | 70.86 | 90.92 | 4.14 |
221
+ | unstructured | 2024-09-26 | 64.49 | 69.90 | 90.42 | 13.14 |
222
+ | google | 2024-09-26 | 62.44 | 68.75 | 90.09 | 5.85 |
223
+ | microsoft | 2024-09-26 | 85.54 | 89.07 | 87.03 | 4.44 |
224
+ | upstage | 2024-09-26 | 91.01 | 93.47 | 96.27 | 3.79 |
225
 
226
  </div>