File size: 11,705 Bytes
f029b51
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1e094a2
f029b51
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1e094a2
 
 
 
 
 
f029b51
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1e094a2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
157579e
f029b51
 
1e094a2
157579e
f029b51
157579e
1e094a2
f029b51
 
 
 
 
 
 
 
157579e
f029b51
 
 
 
 
 
1e094a2
 
f029b51
 
 
 
 
 
 
 
 
 
1e094a2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f029b51
 
 
 
 
 
 
 
 
 
1e094a2
f029b51
 
157579e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f029b51
 
 
 
7461035
756b583
f029b51
 
157579e
f029b51
 
 
 
 
 
 
 
157579e
f029b51
 
 
 
 
 
 
1e094a2
157579e
f029b51
 
157579e
f029b51
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
import ast
import argparse
import glob
import pickle

import gradio as gr
import numpy as np
import pandas as pd
block_css = """
#notice_markdown {
    font-size: 104%
}
#notice_markdown th {
    display: none;
}
#notice_markdown td {
    padding-top: 6px;
    padding-bottom: 6px;
}
#leaderboard_markdown {
    font-size: 104%
}
#leaderboard_markdown td {
    padding-top: 6px;
    padding-bottom: 6px;
}
#leaderboard_dataframe td {
    line-height: 0.1em;
}
footer {
    display:none !important
}
.image-container {
    display: flex;
    align-items: center;
    padding: 1px;
}
.image-container img {
    margin: 0 30px;
    height: 20px;
    max-height: 100%;
    width: auto;
    max-width: 20%;
}
"""
def model_hyperlink(model_name, link):
    return f'<a target="_blank" href="{link}" style="color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;">{model_name}</a>'
def load_leaderboard_table_csv(filename, add_hyperlink=True):
    lines = open(filename).readlines()
    heads = [v.strip() for v in lines[0].split(",")]
    rows = []
    for i in range(1, len(lines)):
        row = [v.strip() for v in lines[i].split(",")]
        for j in range(len(heads)):
            item = {}
            for h, v in zip(heads, row):
                if h != "Model" and h != "Link" and h != "Language Model" and h != "Open Source":
                    item[h] = int(v)
                else:
                    item[h] = v
            if add_hyperlink:
                item["Model"] = model_hyperlink(item["Model"], item["Link"])
        rows.append(item)
    return rows

def get_arena_table(model_table_df):
    # sort by rating
    model_table_df = model_table_df.sort_values(by=["Final Score"], ascending=False)
    values = []
    for i in range(len(model_table_df)):
        row = []
        model_key = model_table_df.index[i]
        model_name = model_table_df["Model"].values[model_key]
        # rank
        row.append(i + 1)
        # model display name
        row.append(model_name)

        row.append(
            model_table_df["Language Model"].values[model_key]
        )
        row.append(
            model_table_df["Open Source"].values[model_key]
        )
        row.append(
            model_table_df["Text Recognition"].values[model_key]
        )

        row.append(
            model_table_df["Scene Text-Centric VQA"].values[model_key]
        )

        row.append(
            model_table_df["Doc-Oriented VQA"].values[model_key]
        )

        row.append(
            model_table_df["KIE"].values[model_key]
        )

        row.append(
            model_table_df["HMER"].values[model_key]
        )

        row.append(
            model_table_df["Final Score"].values[model_key]
        )
        values.append(row)
    return values

def get_recog_table(model_table_df):
    # sort by rating
    values = []
    for i in range(len(model_table_df)):
        row = []
        model_key = model_table_df.index[i]
        model_name = model_table_df["Model"].values[model_key]
        # rank
        row.append(i + 1)
        # model display name
        row.append(model_name)

        row.append(
            model_table_df["Language Model"].values[model_key]
        )
        row.append(
            model_table_df["Open Source"].values[model_key]
        )
        row.append(
            model_table_df["Regular Text"].values[model_key]
        )

        row.append(
            model_table_df["Irregular Text"].values[model_key]
        )

        row.append(
            model_table_df["Artistic Text"].values[model_key]
        )

        row.append(
            model_table_df["Handwriting"].values[model_key]
        )

        row.append(
            model_table_df["Digit string"].values[model_key]
        )

        row.append(
            model_table_df["Non-semantic Text"].values[model_key]
        )
        row.append(
            model_table_df["ALL"].values[model_key]
        )
        values.append(row)
    return values

def build_leaderboard_tab(leaderboard_table_file, text_recog_file, Inaccessible_model_file, show_plot=False):
    if leaderboard_table_file:
        data = load_leaderboard_table_csv(leaderboard_table_file)
        data_recog = load_leaderboard_table_csv(text_recog_file)
        data_Inaccessible = load_leaderboard_table_csv(Inaccessible_model_file)
        model_table_df = pd.DataFrame(data)
        model_table_df_Inaccessible = pd.DataFrame(data_Inaccessible)
        recog_table_df = pd.DataFrame(data_recog)
        md_head = f"""
        # 🏆 OCRBench Leaderboard
        | [GitHub](https://github.com/Yuliang-Liu/MultimodalOCR) | [Paper](https://arxiv.org/abs/2305.07895) |
        """
        gr.Markdown(md_head, elem_id="leaderboard_markdown")
        with gr.Tabs() as tabs:
            # arena table
            with gr.Tab("OCRBench", id=0):
                arena_table_vals = get_arena_table(model_table_df)
                md = "OCRBench is a comprehensive evaluation benchmark designed to assess the OCR capabilities of Large Multimodal Models. It comprises five components: Text Recognition, SceneText-Centric VQA, Document-Oriented VQA, Key Information Extraction, and Handwritten Mathematical Expression Recognition. The benchmark includes 1000 question-answer pairs, and all the answers undergo manual verification and correction to ensure a more precise evaluation."
                gr.Markdown(md, elem_id="leaderboard_markdown")
                gr.Dataframe(
                    headers=[
                        "Rank",
                        "Name",
                        "Language Model",
                        "Open Source",
                        "Text Recognition",
                        "Scene Text-Centric VQA",
                        "Doc-Oriented VQA",
                        "KIE",
                        "HMER",
                        "Final Score",
                    ],
                    datatype=[
                        "str",
                        "markdown",
                        "str",
                        "str",
                        "number",
                        "number",
                        "number",
                        "number",
                        "number",
                        "number",
                    ],
                    value=arena_table_vals,
                    elem_id="arena_leaderboard_dataframe",
                    height=700,
                    column_widths=[60, 120,150,100, 150, 200, 180, 80, 80, 160],
                    wrap=True,
                )
            with gr.Tab("Text Recognition", id=1):
                arena_table_vals = get_recog_table(recog_table_df)
                md = "OCRBench is a comprehensive evaluation benchmark designed to assess the OCR capabilities of Large Multimodal Models. It comprises five components: Text Recognition, SceneText-Centric VQA, Document-Oriented VQA, Key Information Extraction, and Handwritten Mathematical Expression Recognition. The benchmark includes 1000 question-answer pairs, and all the answers undergo manual verification and correction to ensure a more precise evaluation."
                gr.Markdown(md, elem_id="leaderboard_markdown")
                gr.Dataframe(
                    headers=[
                        "Rank",
                        "Name",
                        "Language Model",
                        "Open Source",
                        "Regular Text",
                        "Irregular Text",
                        "Artistic Text",
                        "Handwriting",
                        "Digit string",
                        "Non-semantic Text",
                        "ALL",
                    ],
                    datatype=[
                        "str",
                        "markdown",
                        "str",
                        "str",
                        "number",
                        "number",
                        "number",
                        "number",
                        "number",
                        "number",
                        "number",
                    ],
                    value=arena_table_vals,
                    elem_id="arena_leaderboard_dataframe",
                    height=700,
                    column_widths=[60, 120,150,100, 100, 100, 100, 100, 100,100, 80],
                    wrap=True,
                )
            with gr.Tab("Inaccessible Model", id=2):
                arena_table_vals = get_arena_table(model_table_df_Inaccessible)
                md = "The models on this list are neither open-source nor have API call interfaces available."
                gr.Markdown(md, elem_id="leaderboard_markdown")
                gr.Dataframe(
                    headers=[
                        "Rank",
                        "Name",
                        "Language Model",
                        "Open Source",
                        "Text Recognition",
                        "Scene Text-Centric VQA",
                        "Doc-Oriented VQA",
                        "KIE",
                        "HMER",
                        "Final Score",
                    ],
                    datatype=[
                        "str",
                        "markdown",
                        "str",
                        "str",
                        "number",
                        "number",
                        "number",
                        "number",
                        "number",
                        "number",
                    ],
                    value=arena_table_vals,
                    elem_id="arena_leaderboard_dataframe",
                    height=700,
                    column_widths=[60, 120,150,100, 150, 200, 180, 80, 80, 160],
                    wrap=True,
                )
    else:
        pass
    md_tail = f"""
    # Notice
    Sometimes, API calls to closed-source models may not succeed. In such cases, we will repeat the calls for unsuccessful samples until it becomes impossible to obtain a successful response. It is important to note that due to rigorous security reviews by OpenAI, GPT4V refuses to provide results for the 84 samples in OCRBench.
    If you would like to include your model in the OCRBench leaderboard, please follow the evaluation instructions provided on [GitHub](https://github.com/Yuliang-Liu/MultimodalOCR), [VLMEvalKit](https://github.com/open-compass/VLMEvalKit) or [lmms-eval](https://github.com/EvolvingLMMs-Lab/lmms-eval) and feel free to contact us via email at [email protected]. We will update the leaderboard in time."""
    gr.Markdown(md_tail, elem_id="leaderboard_markdown")

def build_demo(leaderboard_table_file, recog_table_file, Inaccessible_model_file):
    text_size = gr.themes.sizes.text_lg

    with gr.Blocks(
        title="OCRBench Leaderboard",
        theme=gr.themes.Base(text_size=text_size),
        css=block_css,
    ) as demo:
        leader_components = build_leaderboard_tab(
            leaderboard_table_file, recog_table_file,Inaccessible_model_file,show_plot=True
        )
    return demo

if __name__ == "__main__":
    parser = argparse.ArgumentParser()
    parser.add_argument("--share", action="store_true")
    parser.add_argument("--OCRBench_file", type=str, default="./OCRBench.csv")
    parser.add_argument("--TextRecognition_file", type=str, default="./TextRecognition.csv")
    parser.add_argument("--Inaccessible_model_file", type=str, default="./Inaccessible_model.csv")
    args = parser.parse_args()

    demo = build_demo(args.OCRBench_file, args.TextRecognition_file, args.Inaccessible_model_file)
    demo.launch()