aci11 commited on
Commit
35f239a
1 Parent(s): 359ecbf

Create app.py

Browse files
Files changed (1) hide show
  1. app.py +271 -0
app.py ADDED
@@ -0,0 +1,271 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # coding: utf-8
2
+ # [Pix2Text](https://github.com/breezedeus/pix2text): an Open-Source Alternative to Mathpix.
3
+ # Copyright (C) 2022-2024, [Breezedeus](https://www.breezedeus.com).
4
+
5
+ import os
6
+ import json
7
+ import functools
8
+ import random
9
+ import shutil
10
+ import string
11
+ import tempfile
12
+ import time
13
+ import zipfile
14
+ from pathlib import Path
15
+
16
+ import yaml
17
+
18
+ import gradio as gr
19
+ import numpy as np
20
+ from huggingface_hub import hf_hub_download
21
+
22
+ # from cnstd.utils import pil_to_numpy, imsave
23
+
24
+ from pix2text import Pix2Text
25
+ from pix2text.utils import set_logger, merge_line_texts
26
+
27
+ logger = set_logger()
28
+
29
+ LANGUAGES = yaml.safe_load(open('languages.yaml', 'r', encoding='utf-8'))['languages']
30
+ OUTPUT_RESULT_DIR = Path('./output-results')
31
+ OUTPUT_RESULT_DIR.mkdir(exist_ok=True)
32
+
33
+
34
+ def prepare_mfd_model():
35
+ target_fp = './yolov7-model/mfd-yolov7-epoch224-20230613.pt'
36
+ if os.path.exists(target_fp):
37
+ return target_fp
38
+ HF_TOKEN = os.environ.get('HF_TOKEN')
39
+ local_path = hf_hub_download(
40
+ repo_id='breezedeus/paid-models',
41
+ subfolder='cnstd/1.2',
42
+ filename='yolov7-model-20230613.zip',
43
+ repo_type="model",
44
+ cache_dir='./',
45
+ token=HF_TOKEN,
46
+ )
47
+ with zipfile.ZipFile(local_path) as zf:
48
+ zf.extractall('./')
49
+ return target_fp
50
+
51
+
52
+ def get_p2t_model(lan_list: list, mfd_model_name: str, mfr_model_name: str):
53
+ mfd_config = {}
54
+ if 'yolov7_tiny' not in mfd_model_name:
55
+ mfd_fp = prepare_mfd_model()
56
+ mfd_config = dict( # 声明 LayoutAnalyzer 的初始化参数
57
+ model_type='yolov7', # 表示使用的是 YoloV7 模型,而不是 YoloV7_Tiny 模型
58
+ model_fp=mfd_fp, # 注:修改成你的模型文件所存储的路径
59
+ )
60
+ formula_config = {}
61
+ if 'mfr-pro' in mfr_model_name:
62
+ formula_config = dict( # 声明 LayoutAnalyzer 的初始化参数
63
+ model_name='mfr-pro', model_backend='onnx',
64
+ )
65
+ text_formula_config = dict(
66
+ languages=lan_list, mfd=mfd_config, formula=formula_config,
67
+ )
68
+ total_config = {
69
+ 'layout': {'scores_thresh': 0.45},
70
+ 'text_formula': text_formula_config,
71
+ }
72
+ p2t = Pix2Text.from_config(total_configs=total_config,)
73
+ return p2t
74
+
75
+
76
+ def latex_render(latex_str):
77
+ return f"$$\n{latex_str}\n$$"
78
+ # return latex_str
79
+
80
+
81
+ def recognize(
82
+ lang_list, mfd_model_name, mfr_model_name, rec_type, resized_shape, image_file
83
+ ):
84
+ lang_list = [LANGUAGES[l] for l in lang_list]
85
+ p2t = get_p2t_model(lang_list, mfd_model_name, mfr_model_name)
86
+
87
+ # 如果 OUTPUT_RESULT_DIR 文件数量超过 100,按时间删除最早的 100 个文件
88
+ if len(os.listdir(OUTPUT_RESULT_DIR)) > 100:
89
+ shutil.rmtree(OUTPUT_RESULT_DIR)
90
+ OUTPUT_RESULT_DIR.mkdir(exist_ok=True)
91
+
92
+ out_det_fp = './docs/no-det-res.jpg'
93
+ kwargs = dict(
94
+ resized_shape=resized_shape,
95
+ return_text = True,
96
+ auto_line_break = True,
97
+ )
98
+ if rec_type == 'page':
99
+ suffix = list(string.ascii_letters)
100
+ random.shuffle(suffix)
101
+ suffix = ''.join(suffix[:6])
102
+ fp_suffix = f'{time.time()}-{suffix}'
103
+ out_debug_dir = f'out-debug-{fp_suffix}'
104
+ output_dir = OUTPUT_RESULT_DIR / f'output-{fp_suffix}'
105
+ kwargs['save_debug_res'] = OUTPUT_RESULT_DIR / out_debug_dir
106
+ elif rec_type == 'text_formula':
107
+ suffix = list(string.ascii_letters)
108
+ random.shuffle(suffix)
109
+ suffix = ''.join(suffix[:6])
110
+ out_det_fp = f'out-det-{time.time()}-{suffix}.jpg'
111
+ kwargs['save_analysis_res'] = str(OUTPUT_RESULT_DIR / out_det_fp)
112
+
113
+ out = p2t.recognize(image_file, file_type=rec_type, **kwargs)
114
+ out_text = out
115
+ if rec_type == 'page':
116
+ out_text = out.to_markdown(output_dir)
117
+ out_det_fp =kwargs['save_debug_res'] / 'layout_res.jpg'
118
+ elif rec_type == 'text_formula':
119
+ out_det_fp = kwargs['save_analysis_res']
120
+
121
+ return out_text, out_det_fp
122
+
123
+
124
+ def example_func(lang_list, rec_type, image_file):
125
+ return recognize(
126
+ lang_list,
127
+ mfd_model_name='yolov7 (paid)',
128
+ mfr_model_name='mfr-pro (paid)',
129
+ rec_type=rec_type,
130
+ resized_shape=768,
131
+ image_file=image_file,
132
+ )
133
+
134
+
135
+ def main():
136
+ langs = list(LANGUAGES.keys())
137
+ langs.sort(key=lambda x: x.lower())
138
+
139
+ title = ': a Free Alternative to Mathpix'
140
+ examples = [
141
+ [['English'], 'page', 'docs/examples/page.png',],
142
+ [['English'], 'text_formula', 'docs/examples/mixed-en.jpg',],
143
+ [['English', 'Chinese Simplified'], 'text_formula', 'docs/examples/mixed-ch_sim.jpg',],
144
+ [
145
+ ['English', 'Chinese Traditional'],
146
+ 'text_formula',
147
+ 'docs/examples/mixed-ch_tra.jpg',
148
+ ],
149
+ [['English', 'Vietnamese'], 'text_formula', 'docs/examples/mixed-vietnamese.jpg',],
150
+ [['English'], 'formula', 'docs/examples/formula1.png'],
151
+ [['English'], 'formula', 'docs/examples/formula2.jpg'],
152
+ [['English'], 'formula', 'docs/examples/hw-formula.png'],
153
+ [['English', 'Chinese Simplified'], 'text', 'docs/examples/pure-text.jpg',],
154
+ ]
155
+
156
+ table_desc = """
157
+ <div align="center">
158
+ <img src="https://pix2text.readthedocs.io/zh/latest/figs/p2t-logo.png" width="120px"/>
159
+
160
+ [![Visitors](https://api.visitorbadge.io/api/visitors?path=https%3A%2F%2Fhuggingface.co%2Fspaces%2Fbreezedeus%2Fpix2text-demo&labelColor=%23697689&countColor=%23f5c791&style=flat&labelStyle=upper)](https://visitorbadge.io/status?path=https%3A%2F%2Fhuggingface.co%2Fspaces%2Fbreezedeus%2Fpix2text-demo)
161
+
162
+ [![Discord](https://img.shields.io/discord/1200765964434821260?logo=discord&label=Discord)](https://discord.gg/GgD87WM8Tf)
163
+
164
+ | | |
165
+ | ------------------------------- | --------------------------------------- |
166
+ | 🏄 **Online Service** | [p2t.breezedeus.com](https://p2t.breezedeus.com) |
167
+ | 📖 **Doc** | [Online Doc](https://pix2text.readthedocs.io) |
168
+ | 📀 **Code** | [Github](https://github.com/breezedeus/pix2text) |
169
+ | 🤗 **Models** | [breezedeus/Models](https://huggingface.co/breezedeus) |
170
+ | 📄 **More Infos** | [Pix2Text Infos](https://www.breezedeus.com/article/pix2text) |
171
+
172
+ If useful, please help to **star 🌟 [Pix2Text](https://github.com/breezedeus/pix2text)** 🙏
173
+ </div>
174
+ """
175
+
176
+ with gr.Blocks() as demo:
177
+ gr.HTML(
178
+ f'<h1 style="text-align: center; margin-bottom: 1rem;"><a href="https://github.com/breezedeus/pix2text" target="_blank">Pix2Text V1.1</a>{title}</h1>'
179
+ )
180
+ with gr.Row(equal_height=False):
181
+ with gr.Column(min_width=200, variant='panel', scale=3):
182
+ gr.Markdown('### Settings')
183
+ lang_list = gr.Dropdown(
184
+ label='Text Languages',
185
+ choices=langs,
186
+ value=['English', 'Chinese Simplified'],
187
+ multiselect=True,
188
+ # info='Which languages to be recognized as Texts.',
189
+ )
190
+ mfd_model_name = gr.Dropdown(
191
+ label='MFD Models',
192
+ choices=['yolov7_tiny (free)', 'yolov7 (paid)'],
193
+ value='yolov7 (paid)',
194
+ )
195
+ mfr_model_name = gr.Dropdown(
196
+ label='MFR Models',
197
+ choices=['mfr (free)', 'mfr-pro (paid)'],
198
+ value='mfr-pro (paid)',
199
+ )
200
+ rec_type = gr.Dropdown(
201
+ label='File Type',
202
+ choices=['page', 'text_formula', 'formula', 'text'],
203
+ value='text_formula',
204
+ # info='Which type of image to be recognized.',
205
+ )
206
+ with gr.Accordion('More Options', open=False):
207
+ resized_shape = gr.Slider(
208
+ label='resized_shape',
209
+ minimum=512,
210
+ maximum=2048,
211
+ value=768,
212
+ step=32,
213
+ )
214
+
215
+ with gr.Column(scale=6, variant='compact'):
216
+ gr.Markdown('### Upload Image to be Recognized')
217
+ image_file = gr.Image(
218
+ label='Image', type="pil", image_mode='RGB', show_label=False
219
+ )
220
+ sub_btn = gr.Button("Submit", variant="primary")
221
+
222
+ with gr.Column(scale=2, variant='compact'):
223
+ gr.Markdown(table_desc)
224
+ with gr.Row(equal_height=False):
225
+ with gr.Column(scale=1, variant='compact'):
226
+ gr.Markdown('**Detection Result**')
227
+ det_result = gr.Image(
228
+ label='Detection Result', scale=1, show_label=False
229
+ )
230
+ with gr.Column(scale=1, variant='compact'):
231
+ gr.Markdown(
232
+ '**Recognition Results (Paste them into the [P2T Online Service](https://p2t.breezedeus.com) to view rendered outcomes)**'
233
+ )
234
+ rec_result = gr.Textbox(
235
+ label=f'Recognition Result ',
236
+ lines=5,
237
+ value='',
238
+ scale=1,
239
+ show_label=False,
240
+ show_copy_button=True,
241
+ )
242
+ # render_result = gr.Markdown(label=f'After Rendering', value='')
243
+ # rec_result.change(latex_render, rec_result, render_result)
244
+ sub_btn.click(
245
+ recognize,
246
+ inputs=[
247
+ lang_list,
248
+ mfd_model_name,
249
+ mfr_model_name,
250
+ rec_type,
251
+ resized_shape,
252
+ image_file,
253
+ ],
254
+ outputs=[rec_result, det_result],
255
+ )
256
+
257
+ gr.Examples(
258
+ label='Examples',
259
+ examples=examples,
260
+ inputs=[lang_list, rec_type, image_file,],
261
+ outputs=[rec_result, det_result],
262
+ fn=example_func,
263
+ cache_examples=os.getenv('CACHE_EXAMPLES') == '1',
264
+ )
265
+
266
+ demo.queue(max_size=10)
267
+ demo.launch()
268
+
269
+
270
+ if __name__ == '__main__':
271
+ main()