Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
@@ -5,6 +5,8 @@ from transformers import AutoModel, AutoTokenizer
|
|
5 |
import warnings
|
6 |
import os
|
7 |
import spaces
|
|
|
|
|
8 |
|
9 |
# 禁用警告信息
|
10 |
warnings.filterwarnings("ignore")
|
@@ -136,6 +138,199 @@ def clean_text_output(text):
|
|
136 |
|
137 |
return '\n'.join(text_lines)
|
138 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
139 |
@spaces.GPU
|
140 |
def parse_image(image, parse_type):
|
141 |
"""解析图片内容为指定格式"""
|
@@ -175,24 +370,36 @@ def parse_image(image, parse_type):
|
|
175 |
for new_text in res:
|
176 |
generated_text += new_text
|
177 |
|
178 |
-
#
|
179 |
if parse_type == "表格解析":
|
180 |
-
|
181 |
-
|
|
|
|
|
182 |
elif parse_type == "公式解析":
|
183 |
-
|
184 |
-
|
|
|
|
|
185 |
elif parse_type == "文本解析":
|
186 |
-
|
187 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
188 |
else:
|
189 |
result = generated_text.strip()
|
190 |
output_format = "原始输出"
|
191 |
-
|
192 |
-
return result, f"解析完成 - 输出格式: {output_format}"
|
193 |
|
194 |
except Exception as e:
|
195 |
-
|
|
|
196 |
|
197 |
def create_interface():
|
198 |
"""创建Gradio界面"""
|
@@ -206,9 +413,50 @@ def create_interface():
|
|
206 |
font-family: 'Courier New', monospace;
|
207 |
font-size: 14px;
|
208 |
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
209 |
"""
|
210 |
|
211 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
212 |
gr.Markdown("""
|
213 |
# 🚀 MiniCPM 多模态内容解析工具
|
214 |
|
@@ -217,12 +465,12 @@ def create_interface():
|
|
217 |
## 📋 使用说明
|
218 |
1. **上传图片**: 支持 PNG、JPG、JPEG 等格式
|
219 |
2. **选择解析类型**: 根据图片内容选择相应的解析模式
|
220 |
-
3. **获取结果**:
|
221 |
|
222 |
## 🎯 解析类型说明
|
223 |
-
- **📊 表格解析**:
|
224 |
-
- **🧮 公式解析**:
|
225 |
-
- **📝 文本解析**:
|
226 |
""")
|
227 |
|
228 |
with gr.Row():
|
@@ -255,14 +503,20 @@ def create_interface():
|
|
255 |
interactive=False
|
256 |
)
|
257 |
|
258 |
-
result_output = gr.
|
259 |
-
label="📄
|
260 |
-
|
261 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
262 |
show_copy_button=True,
|
263 |
elem_classes=["output-text"],
|
264 |
-
placeholder="
|
265 |
-
|
266 |
)
|
267 |
|
268 |
# 示例图片
|
@@ -283,7 +537,7 @@ def create_interface():
|
|
283 |
parse_button.click(
|
284 |
fn=parse_image,
|
285 |
inputs=[image_input, parse_type],
|
286 |
-
outputs=[result_output, status_output]
|
287 |
)
|
288 |
|
289 |
# 添加页脚信息
|
@@ -294,6 +548,7 @@ def create_interface():
|
|
294 |
- 复杂表格建议分段处理
|
295 |
- 公式图片建议使用高分辨率
|
296 |
- 文字图片避免模糊、倾斜或光线不足
|
|
|
297 |
|
298 |
### 🔧 技术支持
|
299 |
- 模型: MiniCPM-o-2.6
|
|
|
5 |
import warnings
|
6 |
import os
|
7 |
import spaces
|
8 |
+
import markdown
|
9 |
+
import re
|
10 |
|
11 |
# 禁用警告信息
|
12 |
warnings.filterwarnings("ignore")
|
|
|
138 |
|
139 |
return '\n'.join(text_lines)
|
140 |
|
141 |
+
def detect_content_types(text):
|
142 |
+
"""检测文本中包含的内容类型"""
|
143 |
+
content_types = set()
|
144 |
+
|
145 |
+
# 检测表格(Markdown格式)
|
146 |
+
if '|' in text and any(line.count('|') >= 2 for line in text.split('\n')):
|
147 |
+
content_types.add('table')
|
148 |
+
|
149 |
+
# 检测公式(LaTeX格式)
|
150 |
+
formula_indicators = ['$', '\\frac', '\\sum', '\\int', '\\sqrt', '\\alpha', '\\beta', '\\gamma', '\\delta',
|
151 |
+
'\\theta', '\\pi', '\\sigma', '\\omega', '\\infty', '\\partial', '\\nabla']
|
152 |
+
if any(indicator in text for indicator in formula_indicators) or \
|
153 |
+
(any(symbol in text for symbol in ['{', '}', '^', '_']) and any(char.isdigit() for char in text)):
|
154 |
+
content_types.add('formula')
|
155 |
+
|
156 |
+
# 总是包含文本
|
157 |
+
content_types.add('text')
|
158 |
+
|
159 |
+
return content_types
|
160 |
+
|
161 |
+
def render_mixed_content(text):
|
162 |
+
"""渲染混合内容(文本+表格+公式)"""
|
163 |
+
if not text.strip():
|
164 |
+
return text
|
165 |
+
|
166 |
+
# 检测内容类型
|
167 |
+
content_types = detect_content_types(text)
|
168 |
+
|
169 |
+
# 如果只有纯文本,简单处理
|
170 |
+
if content_types == {'text'}:
|
171 |
+
return f"<div style='padding: 15px; white-space: pre-wrap; font-family: Arial, sans-serif; line-height: 1.6;'>{text}</div>"
|
172 |
+
|
173 |
+
# 处理混合内容
|
174 |
+
lines = text.split('\n')
|
175 |
+
rendered_parts = []
|
176 |
+
current_block = []
|
177 |
+
current_type = 'text'
|
178 |
+
|
179 |
+
i = 0
|
180 |
+
while i < len(lines):
|
181 |
+
line = lines[i].strip()
|
182 |
+
|
183 |
+
# 检测表格开始
|
184 |
+
if '|' in line and line.count('|') >= 2:
|
185 |
+
# 先处理之前累积的文本块
|
186 |
+
if current_block and current_type == 'text':
|
187 |
+
text_content = '\n'.join(current_block)
|
188 |
+
if text_content.strip():
|
189 |
+
rendered_parts.append(f"<div style='padding: 10px 0; white-space: pre-wrap; font-family: Arial, sans-serif; line-height: 1.6;'>{text_content}</div>")
|
190 |
+
current_block = []
|
191 |
+
|
192 |
+
# 收集表格行
|
193 |
+
table_lines = []
|
194 |
+
while i < len(lines) and '|' in lines[i]:
|
195 |
+
table_lines.append(lines[i])
|
196 |
+
i += 1
|
197 |
+
|
198 |
+
# 渲染表格
|
199 |
+
if table_lines:
|
200 |
+
table_markdown = '\n'.join(table_lines)
|
201 |
+
table_html = render_markdown_table(table_markdown)
|
202 |
+
rendered_parts.append(table_html)
|
203 |
+
|
204 |
+
current_type = 'text'
|
205 |
+
continue
|
206 |
+
|
207 |
+
# 检测公式(简单检测包含LaTeX符号的行)
|
208 |
+
elif any(symbol in line for symbol in ['$', '\\frac', '\\sum', '\\int', '\\sqrt']) and current_type != 'formula':
|
209 |
+
# 先处理之前累积的文本块
|
210 |
+
if current_block and current_type == 'text':
|
211 |
+
text_content = '\n'.join(current_block)
|
212 |
+
if text_content.strip():
|
213 |
+
rendered_parts.append(f"<div style='padding: 10px 0; white-space: pre-wrap; font-family: Arial, sans-serif; line-height: 1.6;'>{text_content}</div>")
|
214 |
+
current_block = []
|
215 |
+
|
216 |
+
# 收集公式行
|
217 |
+
formula_lines = [line]
|
218 |
+
i += 1
|
219 |
+
while i < len(lines):
|
220 |
+
next_line = lines[i].strip()
|
221 |
+
if any(symbol in next_line for symbol in ['$', '\\', '{', '}', '^', '_']) or \
|
222 |
+
any(char.isdigit() or char in '+-*/=()[]{}^_' for char in next_line):
|
223 |
+
formula_lines.append(next_line)
|
224 |
+
i += 1
|
225 |
+
else:
|
226 |
+
break
|
227 |
+
|
228 |
+
# 渲染公式
|
229 |
+
if formula_lines:
|
230 |
+
formula_text = '\n'.join(formula_lines)
|
231 |
+
formula_html = render_latex_formula(formula_text)
|
232 |
+
rendered_parts.append(formula_html)
|
233 |
+
|
234 |
+
current_type = 'text'
|
235 |
+
continue
|
236 |
+
|
237 |
+
# 普通文本行
|
238 |
+
else:
|
239 |
+
current_block.append(lines[i])
|
240 |
+
current_type = 'text'
|
241 |
+
i += 1
|
242 |
+
|
243 |
+
# 处理最后剩余的文本块
|
244 |
+
if current_block:
|
245 |
+
text_content = '\n'.join(current_block)
|
246 |
+
if text_content.strip():
|
247 |
+
rendered_parts.append(f"<div style='padding: 10px 0; white-space: pre-wrap; font-family: Arial, sans-serif; line-height: 1.6;'>{text_content}</div>")
|
248 |
+
|
249 |
+
# 合并所有渲染部分
|
250 |
+
if rendered_parts:
|
251 |
+
return '<div style="padding: 5px;">' + ''.join(rendered_parts) + '</div>'
|
252 |
+
else:
|
253 |
+
return f"<div style='padding: 15px; white-space: pre-wrap; font-family: Arial, sans-serif; line-height: 1.6;'>{text}</div>"
|
254 |
+
|
255 |
+
def render_markdown_table(markdown_text):
|
256 |
+
"""将Markdown表格转换为HTML渲染格式"""
|
257 |
+
if not markdown_text.strip():
|
258 |
+
return markdown_text
|
259 |
+
|
260 |
+
# 使用markdown库转换为HTML
|
261 |
+
html_content = markdown.markdown(markdown_text, extensions=['tables'])
|
262 |
+
|
263 |
+
# 添加表格样式
|
264 |
+
styled_html = f"""
|
265 |
+
<div style="overflow-x: auto; margin: 10px 0;">
|
266 |
+
<style>
|
267 |
+
table {{
|
268 |
+
border-collapse: collapse;
|
269 |
+
width: 100%;
|
270 |
+
margin: 10px 0;
|
271 |
+
font-family: Arial, sans-serif;
|
272 |
+
}}
|
273 |
+
th, td {{
|
274 |
+
border: 1px solid #ddd;
|
275 |
+
padding: 8px 12px;
|
276 |
+
text-align: left;
|
277 |
+
}}
|
278 |
+
th {{
|
279 |
+
background-color: #f2f2f2;
|
280 |
+
font-weight: bold;
|
281 |
+
}}
|
282 |
+
tr:nth-child(even) {{
|
283 |
+
background-color: #f9f9f9;
|
284 |
+
}}
|
285 |
+
tr:hover {{
|
286 |
+
background-color: #f5f5f5;
|
287 |
+
}}
|
288 |
+
</style>
|
289 |
+
{html_content}
|
290 |
+
</div>
|
291 |
+
"""
|
292 |
+
|
293 |
+
return styled_html
|
294 |
+
|
295 |
+
def render_latex_formula(latex_text):
|
296 |
+
"""将LaTeX公式转换为可渲染的HTML格式"""
|
297 |
+
if not latex_text.strip():
|
298 |
+
return latex_text
|
299 |
+
|
300 |
+
# 处理LaTeX公式,确保正确的MathJax格式
|
301 |
+
lines = latex_text.strip().split('\n')
|
302 |
+
processed_lines = []
|
303 |
+
|
304 |
+
for line in lines:
|
305 |
+
line = line.strip()
|
306 |
+
if line:
|
307 |
+
# 检查是否已经有$符号包围
|
308 |
+
if not (line.startswith('$') and line.endswith('$')):
|
309 |
+
# 如果是单行公式,用$$包围(块级公式)
|
310 |
+
if '=' in line or any(symbol in line for symbol in ['\\', '{', '}', '^', '_']):
|
311 |
+
line = f"$${line}$$"
|
312 |
+
else:
|
313 |
+
line = f"${line}$"
|
314 |
+
processed_lines.append(line)
|
315 |
+
|
316 |
+
formula_html = '<br>'.join(processed_lines)
|
317 |
+
|
318 |
+
# 添加MathJax支持的HTML
|
319 |
+
html_content = f"""
|
320 |
+
<div style="margin: 10px 0; padding: 15px; background-color: #f8f9fa; border-left: 4px solid #007bff; border-radius: 4px;">
|
321 |
+
<div style="font-family: 'Times New Roman', serif; font-size: 16px; line-height: 1.6;">
|
322 |
+
{formula_html}
|
323 |
+
</div>
|
324 |
+
</div>
|
325 |
+
<script>
|
326 |
+
if (typeof MathJax !== 'undefined') {{
|
327 |
+
MathJax.typesetPromise();
|
328 |
+
}}
|
329 |
+
</script>
|
330 |
+
"""
|
331 |
+
|
332 |
+
return html_content
|
333 |
+
|
334 |
@spaces.GPU
|
335 |
def parse_image(image, parse_type):
|
336 |
"""解析图片内容为指定格式"""
|
|
|
370 |
for new_text in res:
|
371 |
generated_text += new_text
|
372 |
|
373 |
+
# 根据类型清理输出并渲染
|
374 |
if parse_type == "表格解析":
|
375 |
+
cleaned_result = clean_markdown_output(generated_text)
|
376 |
+
rendered_result = render_markdown_table(cleaned_result)
|
377 |
+
output_format = "Markdown表格"
|
378 |
+
return rendered_result, cleaned_result, f"解析完成 - 输出格式: {output_format}"
|
379 |
elif parse_type == "公式解析":
|
380 |
+
cleaned_result = clean_formula_output(generated_text)
|
381 |
+
rendered_result = render_latex_formula(cleaned_result)
|
382 |
+
output_format = "LaTeX公式"
|
383 |
+
return rendered_result, cleaned_result, f"解析完成 - 输出格式: {output_format}"
|
384 |
elif parse_type == "文本解析":
|
385 |
+
cleaned_result = clean_text_output(generated_text)
|
386 |
+
# 检测是否包含表格或公式,智能渲染
|
387 |
+
content_types = detect_content_types(cleaned_result)
|
388 |
+
if len(content_types) > 1: # 包含多种内容类型
|
389 |
+
rendered_result = render_mixed_content(cleaned_result)
|
390 |
+
output_format = "混合内容(文本+表格+公式)"
|
391 |
+
else:
|
392 |
+
rendered_result = f"<div style='padding: 15px; white-space: pre-wrap; font-family: Arial, sans-serif; line-height: 1.6;'>{cleaned_result}</div>"
|
393 |
+
output_format = "纯文本"
|
394 |
+
return rendered_result, cleaned_result, f"解析完成 - 输出格式: {output_format}"
|
395 |
else:
|
396 |
result = generated_text.strip()
|
397 |
output_format = "原始输出"
|
398 |
+
return f"<div style='padding: 15px; white-space: pre-wrap; font-family: monospace;'>{result}</div>", result, f"解析完成 - 输出格式: {output_format}"
|
|
|
399 |
|
400 |
except Exception as e:
|
401 |
+
error_html = f"<div style='color: red; padding: 15px; border: 1px solid red; border-radius: 4px;'>解析失败: {str(e)}</div>"
|
402 |
+
return error_html, str(e), "错误"
|
403 |
|
404 |
def create_interface():
|
405 |
"""创建Gradio界面"""
|
|
|
413 |
font-family: 'Courier New', monospace;
|
414 |
font-size: 14px;
|
415 |
}
|
416 |
+
.rendered-output {
|
417 |
+
font-family: Arial, sans-serif;
|
418 |
+
line-height: 1.6;
|
419 |
+
}
|
420 |
+
.rendered-output table {
|
421 |
+
border-collapse: collapse;
|
422 |
+
width: 100%;
|
423 |
+
margin: 10px 0;
|
424 |
+
}
|
425 |
+
.rendered-output th, .rendered-output td {
|
426 |
+
border: 1px solid #ddd;
|
427 |
+
padding: 8px 12px;
|
428 |
+
text-align: left;
|
429 |
+
}
|
430 |
+
.rendered-output th {
|
431 |
+
background-color: #f2f2f2;
|
432 |
+
font-weight: bold;
|
433 |
+
}
|
434 |
+
.rendered-output tr:nth-child(even) {
|
435 |
+
background-color: #f9f9f9;
|
436 |
+
}
|
437 |
"""
|
438 |
|
439 |
+
# MathJax配置
|
440 |
+
mathjax_config = """
|
441 |
+
<script>
|
442 |
+
window.MathJax = {
|
443 |
+
tex: {
|
444 |
+
inlineMath: [['$', '$'], ['\\(', '\\)']],
|
445 |
+
displayMath: [['$$', '$$'], ['\\[', '\\]']],
|
446 |
+
processEscapes: true,
|
447 |
+
processEnvironments: true
|
448 |
+
},
|
449 |
+
options: {
|
450 |
+
ignoreHtmlClass: 'tex2jax_ignore',
|
451 |
+
processHtmlClass: 'tex2jax_process'
|
452 |
+
}
|
453 |
+
};
|
454 |
+
</script>
|
455 |
+
<script src="https://polyfill.io/v3/polyfill.min.js?features=es6"></script>
|
456 |
+
<script id="MathJax-script" async src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-mml-chtml.js"></script>
|
457 |
+
"""
|
458 |
+
|
459 |
+
with gr.Blocks(css=css, title="MiniCPM 多模态内容解析工具", analytics_enabled=False, head=mathjax_config) as interface:
|
460 |
gr.Markdown("""
|
461 |
# 🚀 MiniCPM 多模态内容解析工具
|
462 |
|
|
|
465 |
## 📋 使用说明
|
466 |
1. **上传图片**: 支持 PNG、JPG、JPEG 等格式
|
467 |
2. **选择解析类型**: 根据图片内容选择相应的解析模式
|
468 |
+
3. **获取结果**: 自动渲染显示,表格和公式直接可视化
|
469 |
|
470 |
## 🎯 解析类型说明
|
471 |
+
- **📊 表格解析**: 将表格图片转换为可视化表格
|
472 |
+
- **🧮 公式解析**: 识别数学公式并渲染显示
|
473 |
+
- **📝 文本解析**: 提取图片中的所有文字内容,智能识别并渲染其中的表格和公式
|
474 |
""")
|
475 |
|
476 |
with gr.Row():
|
|
|
503 |
interactive=False
|
504 |
)
|
505 |
|
506 |
+
result_output = gr.HTML(
|
507 |
+
label="📄 解析结果(渲染视图)",
|
508 |
+
value="<p style='color: #666; text-align: center; padding: 20px;'>解析结果将在这里显示...</p>",
|
509 |
+
elem_classes=["rendered-output"]
|
510 |
+
)
|
511 |
+
|
512 |
+
raw_output = gr.Textbox(
|
513 |
+
label="📝 原始代码(可复制)",
|
514 |
+
lines=8,
|
515 |
+
max_lines=15,
|
516 |
show_copy_button=True,
|
517 |
elem_classes=["output-text"],
|
518 |
+
placeholder="原始Markdown/LaTeX代码将在这里显示...",
|
519 |
+
visible=False
|
520 |
)
|
521 |
|
522 |
# 示例图片
|
|
|
537 |
parse_button.click(
|
538 |
fn=parse_image,
|
539 |
inputs=[image_input, parse_type],
|
540 |
+
outputs=[result_output, raw_output, status_output]
|
541 |
)
|
542 |
|
543 |
# 添加页脚信息
|
|
|
548 |
- 复杂表格建议分段处理
|
549 |
- 公式图片建议使用高分辨率
|
550 |
- 文字图片避免模糊、倾斜或光线不足
|
551 |
+
- **文本解析**现在支持智能识别:如果文本中包含表格或公式,会自动渲染显示
|
552 |
|
553 |
### 🔧 技术支持
|
554 |
- 模型: MiniCPM-o-2.6
|