Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -7,170 +7,135 @@ from pathlib import Path
|
|
7 |
import pycountry
|
8 |
import json
|
9 |
import logging
|
10 |
-
from tenacity import retry, stop_after_attempt,
|
11 |
import tempfile
|
12 |
-
from typing import Union, Dict, List
|
13 |
from contextlib import contextmanager
|
14 |
import requests
|
15 |
import shutil
|
|
|
|
|
16 |
|
17 |
# Constants
|
18 |
DEFAULT_LANGUAGE = "English"
|
19 |
-
SUPPORTED_IMAGE_TYPES = [".jpg", ".png"]
|
20 |
SUPPORTED_PDF_TYPES = [".pdf"]
|
21 |
TEMP_FILE_EXPIRY = 7200 # 2 hours in seconds
|
22 |
-
UPLOAD_FOLDER = "uploads"
|
|
|
|
|
23 |
|
24 |
-
#
|
25 |
os.makedirs(UPLOAD_FOLDER, exist_ok=True)
|
26 |
-
|
27 |
-
|
28 |
-
|
|
|
|
|
29 |
logger = logging.getLogger(__name__)
|
30 |
|
31 |
class OCRProcessor:
|
32 |
def __init__(self, api_key: str):
|
33 |
-
|
34 |
-
raise ValueError("API key must be provided")
|
35 |
-
self.api_key = api_key
|
36 |
self.client = Mistral(api_key=self.api_key)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
37 |
try:
|
38 |
-
models = self.client.models.list()
|
39 |
if not models:
|
40 |
raise ValueError("No models available")
|
41 |
except Exception as e:
|
42 |
-
raise ValueError(f"
|
43 |
|
44 |
@staticmethod
|
45 |
-
def
|
46 |
-
|
47 |
-
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
|
52 |
-
|
53 |
-
|
54 |
-
|
|
|
|
|
|
|
|
|
|
|
55 |
|
56 |
@staticmethod
|
57 |
def _save_uploaded_file(file_input: Union[str, bytes], filename: str) -> str:
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
|
|
|
67 |
else:
|
68 |
-
|
69 |
-
|
70 |
-
shutil.copy2(file_input, file_path)
|
71 |
-
else:
|
72 |
-
return file_input # Return original path if same file
|
73 |
-
else:
|
74 |
-
with open(file_path, 'wb') as f:
|
75 |
-
if hasattr(file_input, 'read'):
|
76 |
-
shutil.copyfileobj(file_input, f)
|
77 |
-
else:
|
78 |
-
f.write(file_input)
|
79 |
-
return file_path
|
80 |
-
except Exception as e:
|
81 |
-
logger.error(f"Error saving file: {str(e)}")
|
82 |
-
return None
|
83 |
|
84 |
@staticmethod
|
85 |
def _pdf_to_images(pdf_path: str) -> List[str]:
|
86 |
-
|
87 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
88 |
try:
|
89 |
pdf_document = fitz.open(pdf_path)
|
90 |
-
|
91 |
-
|
92 |
-
|
93 |
-
|
94 |
-
pix.save(image_path)
|
95 |
-
image_paths.append(image_path)
|
96 |
pdf_document.close()
|
97 |
-
return
|
98 |
except Exception as e:
|
99 |
-
logger.error(f"Error converting
|
100 |
-
return
|
101 |
|
102 |
-
@
|
103 |
-
@contextmanager
|
104 |
-
def _temp_file(content: bytes, suffix: str) -> str:
|
105 |
-
temp_file = tempfile.NamedTemporaryFile(delete=False, suffix=suffix)
|
106 |
-
try:
|
107 |
-
temp_file.write(content)
|
108 |
-
temp_file.close()
|
109 |
-
yield temp_file.name
|
110 |
-
finally:
|
111 |
-
if os.path.exists(temp_file.name):
|
112 |
-
os.unlink(temp_file.name)
|
113 |
-
|
114 |
-
@retry(stop=stop_after_attempt(3), wait=wait_fixed(2))
|
115 |
def _call_ocr_api(self, document: Union[DocumentURLChunk, ImageURLChunk]) -> OCRResponse:
|
116 |
-
|
117 |
-
|
118 |
-
|
119 |
-
|
120 |
-
|
121 |
|
122 |
-
@retry(stop=stop_after_attempt(3), wait=
|
123 |
def _call_chat_complete(self, model: str, messages: List[Dict], **kwargs) -> Dict:
|
124 |
-
|
125 |
-
return self.client.chat.complete(model=model, messages=messages, **kwargs)
|
126 |
-
except Exception as e:
|
127 |
-
logger.error(f"Chat complete API call failed: {str(e)}")
|
128 |
-
raise
|
129 |
-
|
130 |
-
def _get_file_content(self, file_input: Union[str, bytes]) -> bytes:
|
131 |
-
if isinstance(file_input, str):
|
132 |
-
if file_input.startswith("http"):
|
133 |
-
response = requests.get(file_input)
|
134 |
-
response.raise_for_status()
|
135 |
-
return response.content
|
136 |
-
else:
|
137 |
-
with open(file_input, "rb") as f:
|
138 |
-
return f.read()
|
139 |
-
return file_input.read() if hasattr(file_input, 'read') else file_input
|
140 |
-
|
141 |
-
def ocr_pdf_url(self, pdf_url: str) -> tuple[str, List[str]]:
|
142 |
-
logger.info(f"Processing PDF URL: {pdf_url}")
|
143 |
-
try:
|
144 |
-
# Download and save PDF
|
145 |
-
response = requests.get(pdf_url)
|
146 |
-
response.raise_for_status()
|
147 |
-
filename = pdf_url.split('/')[-1]
|
148 |
-
pdf_path = self._save_uploaded_file(response.content, filename)
|
149 |
-
if not pdf_path:
|
150 |
-
return self._handle_error("PDF saving", Exception("Failed to save PDF")), []
|
151 |
-
|
152 |
-
# Convert PDF to images for visualization
|
153 |
-
image_paths = self._pdf_to_images(pdf_path)
|
154 |
-
|
155 |
-
# Process with OCR
|
156 |
-
response = self._call_ocr_api(DocumentURLChunk(document_url=pdf_url))
|
157 |
-
return self._get_combined_markdown(response), image_paths
|
158 |
-
except Exception as e:
|
159 |
-
return self._handle_error("PDF URL processing", e), []
|
160 |
|
161 |
-
def ocr_uploaded_pdf(self, pdf_file: Union[str, bytes]) ->
|
162 |
-
file_name = getattr(pdf_file, 'name',
|
163 |
logger.info(f"Processing uploaded PDF: {file_name}")
|
164 |
try:
|
165 |
-
|
166 |
pdf_path = self._save_uploaded_file(pdf_file, file_name)
|
167 |
-
if not pdf_path:
|
168 |
-
return self._handle_error("PDF saving", Exception("Failed to save PDF")), []
|
169 |
-
|
170 |
-
# Convert PDF to images for visualization
|
171 |
image_paths = self._pdf_to_images(pdf_path)
|
172 |
|
173 |
-
# Process with OCR
|
174 |
uploaded_file = self.client.files.upload(
|
175 |
file={"file_name": pdf_path, "content": open(pdf_path, "rb")},
|
176 |
purpose="ocr"
|
@@ -179,184 +144,189 @@ class OCRProcessor:
|
|
179 |
response = self._call_ocr_api(DocumentURLChunk(document_url=signed_url.url))
|
180 |
return self._get_combined_markdown(response), image_paths
|
181 |
except Exception as e:
|
182 |
-
return self._handle_error("
|
183 |
|
184 |
-
def
|
185 |
-
|
186 |
-
try:
|
187 |
-
# Download and save image
|
188 |
-
response = requests.get(image_url)
|
189 |
-
response.raise_for_status()
|
190 |
-
filename = image_url.split('/')[-1]
|
191 |
-
image_path = self._save_uploaded_file(response.content, filename)
|
192 |
-
if not image_path:
|
193 |
-
return self._handle_error("image saving", Exception("Failed to save image")), None
|
194 |
-
|
195 |
-
# Process with OCR
|
196 |
-
response = self._call_ocr_api(ImageURLChunk(image_url=image_url))
|
197 |
-
return self._get_combined_markdown(response), image_path
|
198 |
-
except Exception as e:
|
199 |
-
return self._handle_error("image URL processing", e), None
|
200 |
-
|
201 |
-
def ocr_uploaded_image(self, image_file: Union[str, bytes]) -> tuple[str, str]:
|
202 |
-
file_name = getattr(image_file, 'name', 'unknown')
|
203 |
logger.info(f"Processing uploaded image: {file_name}")
|
204 |
try:
|
205 |
-
|
206 |
image_path = self._save_uploaded_file(image_file, file_name)
|
207 |
-
if not image_path:
|
208 |
-
return self._handle_error("image saving", Exception("Failed to save image")), None
|
209 |
-
|
210 |
-
# Process with OCR
|
211 |
encoded_image = self._encode_image(image_path)
|
212 |
-
if encoded_image is None:
|
213 |
-
return self._handle_error("image encoding", Exception("Failed to encode image")), None
|
214 |
base64_url = f"data:image/jpeg;base64,{encoded_image}"
|
215 |
response = self._call_ocr_api(ImageURLChunk(image_url=base64_url))
|
216 |
return self._get_combined_markdown(response), image_path
|
217 |
except Exception as e:
|
218 |
-
return self._handle_error("
|
219 |
|
220 |
def document_understanding(self, doc_url: str, question: str) -> str:
|
221 |
-
logger.info(f"Document understanding - URL: {doc_url}, Question: {question}")
|
222 |
try:
|
223 |
messages = [{"role": "user", "content": [
|
224 |
TextChunk(text=question),
|
225 |
DocumentURLChunk(document_url=doc_url)
|
226 |
]}]
|
227 |
-
response = self._call_chat_complete(
|
228 |
-
|
|
|
|
|
|
|
|
|
229 |
except Exception as e:
|
230 |
return self._handle_error("document understanding", e)
|
231 |
|
232 |
-
def structured_ocr(self, image_file: Union[str, bytes]) ->
|
233 |
-
file_name = getattr(image_file, 'name',
|
234 |
-
logger.info(f"Processing structured OCR for: {file_name}")
|
235 |
try:
|
236 |
-
|
237 |
image_path = self._save_uploaded_file(image_file, file_name)
|
238 |
-
if not image_path:
|
239 |
-
return self._handle_error("image saving", Exception("Failed to save image")), None
|
240 |
-
|
241 |
encoded_image = self._encode_image(image_path)
|
242 |
-
if encoded_image is None:
|
243 |
-
return self._handle_error("image encoding", Exception("Failed to encode image")), None
|
244 |
base64_url = f"data:image/jpeg;base64,{encoded_image}"
|
|
|
245 |
ocr_response = self._call_ocr_api(ImageURLChunk(image_url=base64_url))
|
246 |
markdown = self._get_combined_markdown(ocr_response)
|
247 |
|
248 |
chat_response = self._call_chat_complete(
|
249 |
model="pixtral-12b-latest",
|
250 |
messages=[{
|
251 |
-
"role": "user",
|
252 |
"content": [
|
253 |
ImageURLChunk(image_url=base64_url),
|
254 |
TextChunk(text=(
|
255 |
f"This is image's OCR in markdown:\n<BEGIN_IMAGE_OCR>\n{markdown}\n<END_IMAGE_OCR>.\n"
|
256 |
-
"Convert this into a
|
257 |
))
|
258 |
]
|
259 |
}],
|
260 |
response_format={"type": "json_object"},
|
261 |
-
temperature=0
|
262 |
)
|
263 |
-
|
264 |
-
response_content = chat_response.choices[0].message.content
|
265 |
-
content = json.loads(response_content)
|
266 |
-
return self._format_structured_response(image_path, content), image_path
|
267 |
except Exception as e:
|
268 |
return self._handle_error("structured OCR", e), None
|
269 |
|
270 |
-
|
271 |
-
|
272 |
-
|
273 |
-
|
274 |
-
|
275 |
-
|
276 |
-
markdown = page.markdown
|
277 |
-
for img_name, base64_str in image_data.items():
|
278 |
-
markdown = markdown.replace(f"", f"")
|
279 |
-
markdowns.append(markdown)
|
280 |
-
return "\n\n".join(markdowns)
|
281 |
|
282 |
@staticmethod
|
283 |
def _handle_error(context: str, error: Exception) -> str:
|
284 |
logger.error(f"Error in {context}: {str(error)}")
|
285 |
-
return f"**Error:** {str(error)}"
|
286 |
|
287 |
@staticmethod
|
288 |
def _format_structured_response(file_path: str, content: Dict) -> str:
|
289 |
languages = {lang.alpha_2: lang.name for lang in pycountry.languages if hasattr(lang, 'alpha_2')}
|
290 |
-
|
291 |
-
|
292 |
-
valid_langs = [l for l in content_languages if l in languages.values()]
|
293 |
|
294 |
response = {
|
295 |
"file_name": Path(file_path).name,
|
296 |
-
"topics": content
|
297 |
-
"languages": valid_langs
|
298 |
-
"ocr_contents": content
|
299 |
}
|
300 |
-
return f"```json\n{json.dumps(response, indent=
|
301 |
|
302 |
def create_interface():
|
303 |
-
|
304 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
305 |
|
306 |
-
api_key = gr.Textbox(label="API Key", type="password")
|
307 |
processor_state = gr.State()
|
308 |
-
status = gr.Markdown()
|
309 |
|
310 |
def init_processor(key):
|
311 |
try:
|
312 |
processor = OCRProcessor(key)
|
313 |
-
return processor, "API key validated
|
314 |
except Exception as e:
|
315 |
-
return None, f"Error: {str(e)}"
|
316 |
|
317 |
-
|
318 |
fn=init_processor,
|
319 |
inputs=api_key,
|
320 |
outputs=[processor_state, status]
|
321 |
)
|
322 |
|
323 |
with gr.Tab("Image OCR"):
|
324 |
-
|
325 |
-
|
326 |
-
|
|
|
|
|
|
|
|
|
|
|
327 |
|
328 |
def process_image(processor, image):
|
329 |
-
if not processor:
|
330 |
-
return "Please set API key
|
331 |
-
|
332 |
-
return ocr_result, image_path
|
333 |
|
334 |
-
|
335 |
fn=process_image,
|
336 |
inputs=[processor_state, image_input],
|
337 |
outputs=[image_output, image_preview]
|
338 |
)
|
339 |
|
340 |
with gr.Tab("PDF OCR"):
|
341 |
-
|
342 |
-
|
343 |
-
|
|
|
|
|
|
|
|
|
|
|
344 |
|
345 |
def process_pdf(processor, pdf):
|
346 |
-
if not processor:
|
347 |
-
return "Please set API key
|
348 |
-
|
349 |
-
return ocr_result, image_paths
|
350 |
|
351 |
-
|
352 |
fn=process_pdf,
|
353 |
inputs=[processor_state, pdf_input],
|
354 |
outputs=[pdf_output, pdf_gallery]
|
355 |
)
|
356 |
|
357 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
358 |
|
|
|
359 |
|
360 |
if __name__ == "__main__":
|
361 |
-
|
362 |
-
|
|
|
|
|
|
|
|
|
|
7 |
import pycountry
|
8 |
import json
|
9 |
import logging
|
10 |
+
from tenacity import retry, stop_after_attempt, wait_exponential
|
11 |
import tempfile
|
12 |
+
from typing import Union, Dict, List, Optional, Tuple
|
13 |
from contextlib import contextmanager
|
14 |
import requests
|
15 |
import shutil
|
16 |
+
from concurrent.futures import ThreadPoolExecutor
|
17 |
+
import time
|
18 |
|
19 |
# Constants
|
20 |
DEFAULT_LANGUAGE = "English"
|
21 |
+
SUPPORTED_IMAGE_TYPES = [".jpg", ".png", ".jpeg"]
|
22 |
SUPPORTED_PDF_TYPES = [".pdf"]
|
23 |
TEMP_FILE_EXPIRY = 7200 # 2 hours in seconds
|
24 |
+
UPLOAD_FOLDER = "uploads"
|
25 |
+
MAX_FILE_SIZE = 50 * 1024 * 1024 # 50MB
|
26 |
+
MAX_PDF_PAGES = 50
|
27 |
|
28 |
+
# Configuration
|
29 |
os.makedirs(UPLOAD_FOLDER, exist_ok=True)
|
30 |
+
logging.basicConfig(
|
31 |
+
level=logging.INFO,
|
32 |
+
format='%(asctime)s - %(name)s - %(levelname)s - %(message)s',
|
33 |
+
handlers=[logging.StreamHandler()]
|
34 |
+
)
|
35 |
logger = logging.getLogger(__name__)
|
36 |
|
37 |
class OCRProcessor:
|
38 |
def __init__(self, api_key: str):
|
39 |
+
self.api_key = self._validate_api_key(api_key)
|
|
|
|
|
40 |
self.client = Mistral(api_key=self.api_key)
|
41 |
+
self._validate_client()
|
42 |
+
|
43 |
+
@staticmethod
|
44 |
+
def _validate_api_key(api_key: str) -> str:
|
45 |
+
if not api_key or not isinstance(api_key, str):
|
46 |
+
raise ValueError("Valid API key must be provided")
|
47 |
+
return api_key
|
48 |
+
|
49 |
+
def _validate_client(self) -> None:
|
50 |
try:
|
51 |
+
models = self.client.models.list()
|
52 |
if not models:
|
53 |
raise ValueError("No models available")
|
54 |
except Exception as e:
|
55 |
+
raise ValueError(f"API key validation failed: {str(e)}")
|
56 |
|
57 |
@staticmethod
|
58 |
+
def _check_file_size(file_input: Union[str, bytes]) -> None:
|
59 |
+
if isinstance(file_input, str) and os.path.exists(file_input):
|
60 |
+
size = os.path.getsize(file_input)
|
61 |
+
elif hasattr(file_input, 'read'):
|
62 |
+
size = len(file_input.read())
|
63 |
+
file_input.seek(0) # Reset file pointer
|
64 |
+
else:
|
65 |
+
size = len(file_input)
|
66 |
+
if size > MAX_FILE_SIZE:
|
67 |
+
raise ValueError(f"File size exceeds {MAX_FILE_SIZE/1024/1024}MB limit")
|
68 |
+
|
69 |
+
@staticmethod
|
70 |
+
def _encode_image(image_path: str) -> Optional[str]:
|
71 |
+
with open(image_path, "rb") as image_file:
|
72 |
+
return base64.b64encode(image_file.read()).decode('utf-8')
|
73 |
|
74 |
@staticmethod
|
75 |
def _save_uploaded_file(file_input: Union[str, bytes], filename: str) -> str:
|
76 |
+
file_path = os.path.join(UPLOAD_FOLDER, f"{int(time.time())}_{filename}")
|
77 |
+
if isinstance(file_input, str) and file_input.startswith("http"):
|
78 |
+
response = requests.get(file_input, timeout=10)
|
79 |
+
response.raise_for_status()
|
80 |
+
with open(file_path, 'wb') as f:
|
81 |
+
f.write(response.content)
|
82 |
+
else:
|
83 |
+
with open(file_path, 'wb') as f:
|
84 |
+
if hasattr(file_input, 'read'):
|
85 |
+
shutil.copyfileobj(file_input, f)
|
86 |
else:
|
87 |
+
f.write(file_input)
|
88 |
+
return file_path
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
89 |
|
90 |
@staticmethod
|
91 |
def _pdf_to_images(pdf_path: str) -> List[str]:
|
92 |
+
pdf_document = fitz.open(pdf_path)
|
93 |
+
if pdf_document.page_count > MAX_PDF_PAGES:
|
94 |
+
pdf_document.close()
|
95 |
+
raise ValueError(f"PDF exceeds maximum page limit of {MAX_PDF_PAGES}")
|
96 |
+
|
97 |
+
with ThreadPoolExecutor() as executor:
|
98 |
+
image_paths = list(executor.map(
|
99 |
+
lambda i: OCRProcessor._convert_page(pdf_path, i),
|
100 |
+
range(pdf_document.page_count)
|
101 |
+
))
|
102 |
+
pdf_document.close()
|
103 |
+
return [path for path in image_paths if path]
|
104 |
+
|
105 |
+
@staticmethod
|
106 |
+
def _convert_page(pdf_path: str, page_num: int) -> Optional[str]:
|
107 |
try:
|
108 |
pdf_document = fitz.open(pdf_path)
|
109 |
+
page = pdf_document[page_num]
|
110 |
+
pix = page.get_pixmap(dpi=150) # Improved resolution
|
111 |
+
image_path = os.path.join(UPLOAD_FOLDER, f"page_{page_num + 1}_{int(time.time())}.png")
|
112 |
+
pix.save(image_path)
|
|
|
|
|
113 |
pdf_document.close()
|
114 |
+
return image_path
|
115 |
except Exception as e:
|
116 |
+
logger.error(f"Error converting page {page_num}: {str(e)}")
|
117 |
+
return None
|
118 |
|
119 |
+
@retry(stop=stop_after_attempt(3), wait=wait_exponential(multiplier=1, min=2, max=10))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
120 |
def _call_ocr_api(self, document: Union[DocumentURLChunk, ImageURLChunk]) -> OCRResponse:
|
121 |
+
return self.client.ocr.process(
|
122 |
+
model="mistral-ocr-latest",
|
123 |
+
document=document,
|
124 |
+
include_image_base64=True
|
125 |
+
)
|
126 |
|
127 |
+
@retry(stop=stop_after_attempt(3), wait=wait_exponential(multiplier=1, min=2, max=10))
|
128 |
def _call_chat_complete(self, model: str, messages: List[Dict], **kwargs) -> Dict:
|
129 |
+
return self.client.chat.complete(model=model, messages=messages, **kwargs)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
130 |
|
131 |
+
def ocr_uploaded_pdf(self, pdf_file: Union[str, bytes]) -> Tuple[str, List[str]]:
|
132 |
+
file_name = getattr(pdf_file, 'name', f"pdf_{int(time.time())}.pdf")
|
133 |
logger.info(f"Processing uploaded PDF: {file_name}")
|
134 |
try:
|
135 |
+
self._check_file_size(pdf_file)
|
136 |
pdf_path = self._save_uploaded_file(pdf_file, file_name)
|
|
|
|
|
|
|
|
|
137 |
image_paths = self._pdf_to_images(pdf_path)
|
138 |
|
|
|
139 |
uploaded_file = self.client.files.upload(
|
140 |
file={"file_name": pdf_path, "content": open(pdf_path, "rb")},
|
141 |
purpose="ocr"
|
|
|
144 |
response = self._call_ocr_api(DocumentURLChunk(document_url=signed_url.url))
|
145 |
return self._get_combined_markdown(response), image_paths
|
146 |
except Exception as e:
|
147 |
+
return self._handle_error("PDF processing", e), []
|
148 |
|
149 |
+
def ocr_uploaded_image(self, image_file: Union[str, bytes]) -> Tuple[str, str]:
|
150 |
+
file_name = getattr(image_file, 'name', f"image_{int(time.time())}.jpg")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
151 |
logger.info(f"Processing uploaded image: {file_name}")
|
152 |
try:
|
153 |
+
self._check_file_size(image_file)
|
154 |
image_path = self._save_uploaded_file(image_file, file_name)
|
|
|
|
|
|
|
|
|
155 |
encoded_image = self._encode_image(image_path)
|
|
|
|
|
156 |
base64_url = f"data:image/jpeg;base64,{encoded_image}"
|
157 |
response = self._call_ocr_api(ImageURLChunk(image_url=base64_url))
|
158 |
return self._get_combined_markdown(response), image_path
|
159 |
except Exception as e:
|
160 |
+
return self._handle_error("image processing", e), None
|
161 |
|
162 |
def document_understanding(self, doc_url: str, question: str) -> str:
|
|
|
163 |
try:
|
164 |
messages = [{"role": "user", "content": [
|
165 |
TextChunk(text=question),
|
166 |
DocumentURLChunk(document_url=doc_url)
|
167 |
]}]
|
168 |
+
response = self._call_chat_complete(
|
169 |
+
model="mistral-small-latest",
|
170 |
+
messages=messages,
|
171 |
+
temperature=0.1
|
172 |
+
)
|
173 |
+
return response.choices[0].message.content
|
174 |
except Exception as e:
|
175 |
return self._handle_error("document understanding", e)
|
176 |
|
177 |
+
def structured_ocr(self, image_file: Union[str, bytes]) -> Tuple[str, str]:
|
178 |
+
file_name = getattr(image_file, 'name', f"image_{int(time.time())}.jpg")
|
|
|
179 |
try:
|
180 |
+
self._check_file_size(image_file)
|
181 |
image_path = self._save_uploaded_file(image_file, file_name)
|
|
|
|
|
|
|
182 |
encoded_image = self._encode_image(image_path)
|
|
|
|
|
183 |
base64_url = f"data:image/jpeg;base64,{encoded_image}"
|
184 |
+
|
185 |
ocr_response = self._call_ocr_api(ImageURLChunk(image_url=base64_url))
|
186 |
markdown = self._get_combined_markdown(ocr_response)
|
187 |
|
188 |
chat_response = self._call_chat_complete(
|
189 |
model="pixtral-12b-latest",
|
190 |
messages=[{
|
191 |
+
"role": "user",
|
192 |
"content": [
|
193 |
ImageURLChunk(image_url=base64_url),
|
194 |
TextChunk(text=(
|
195 |
f"This is image's OCR in markdown:\n<BEGIN_IMAGE_OCR>\n{markdown}\n<END_IMAGE_OCR>.\n"
|
196 |
+
"Convert this into a structured JSON response with file_name, topics, languages, and ocr_contents fields"
|
197 |
))
|
198 |
]
|
199 |
}],
|
200 |
response_format={"type": "json_object"},
|
201 |
+
temperature=0.1
|
202 |
)
|
203 |
+
return self._format_structured_response(image_path, json.loads(chat_response.choices[0].message.content)), image_path
|
|
|
|
|
|
|
204 |
except Exception as e:
|
205 |
return self._handle_error("structured OCR", e), None
|
206 |
|
207 |
+
@staticmethod
|
208 |
+
def _get_combined_markdown(response: OCRResponse) -> str:
|
209 |
+
return "\n\n".join(
|
210 |
+
page.markdown for page in response.pages
|
211 |
+
if page.markdown.strip()
|
212 |
+
) or "No text detected"
|
|
|
|
|
|
|
|
|
|
|
213 |
|
214 |
@staticmethod
|
215 |
def _handle_error(context: str, error: Exception) -> str:
|
216 |
logger.error(f"Error in {context}: {str(error)}")
|
217 |
+
return f"**Error in {context}:** {str(error)}"
|
218 |
|
219 |
@staticmethod
|
220 |
def _format_structured_response(file_path: str, content: Dict) -> str:
|
221 |
languages = {lang.alpha_2: lang.name for lang in pycountry.languages if hasattr(lang, 'alpha_2')}
|
222 |
+
content_languages = content.get("languages", [DEFAULT_LANGUAGE])
|
223 |
+
valid_langs = [l for l in content_languages if l in languages.values()] or [DEFAULT_LANGUAGE]
|
|
|
224 |
|
225 |
response = {
|
226 |
"file_name": Path(file_path).name,
|
227 |
+
"topics": content.get("topics", []),
|
228 |
+
"languages": valid_langs,
|
229 |
+
"ocr_contents": content.get("ocr_contents", {})
|
230 |
}
|
231 |
+
return f"```json\n{json.dumps(response, indent=2, ensure_ascii=False)}\n```"
|
232 |
|
233 |
def create_interface():
|
234 |
+
css = """
|
235 |
+
.output-markdown {font-size: 14px; max-height: 500px; overflow-y: auto;}
|
236 |
+
.status {color: #666; font-style: italic;}
|
237 |
+
"""
|
238 |
+
|
239 |
+
with gr.Blocks(title="Mistral OCR App", css=css) as demo:
|
240 |
+
gr.Markdown("# Mistral OCR App\nUpload images or PDFs for OCR processing")
|
241 |
+
|
242 |
+
with gr.Row():
|
243 |
+
api_key = gr.Textbox(label="Mistral API Key", type="password", placeholder="Enter your API key")
|
244 |
+
set_key_btn = gr.Button("Set API Key", variant="primary")
|
245 |
|
|
|
246 |
processor_state = gr.State()
|
247 |
+
status = gr.Markdown("Please enter API key", elem_classes="status")
|
248 |
|
249 |
def init_processor(key):
|
250 |
try:
|
251 |
processor = OCRProcessor(key)
|
252 |
+
return processor, "✅ API key validated successfully"
|
253 |
except Exception as e:
|
254 |
+
return None, f"❌ Error: {str(e)}"
|
255 |
|
256 |
+
set_key_btn.click(
|
257 |
fn=init_processor,
|
258 |
inputs=api_key,
|
259 |
outputs=[processor_state, status]
|
260 |
)
|
261 |
|
262 |
with gr.Tab("Image OCR"):
|
263 |
+
with gr.Row():
|
264 |
+
image_input = gr.File(
|
265 |
+
label=f"Upload Image (max {MAX_FILE_SIZE/1024/1024}MB)",
|
266 |
+
file_types=SUPPORTED_IMAGE_TYPES
|
267 |
+
)
|
268 |
+
image_preview = gr.Image(label="Preview", height=300)
|
269 |
+
image_output = gr.Markdown(label="OCR Result", elem_classes="output-markdown")
|
270 |
+
process_image_btn = gr.Button("Process Image", variant="primary")
|
271 |
|
272 |
def process_image(processor, image):
|
273 |
+
if not processor or not image:
|
274 |
+
return "Please set API key and upload an image", None
|
275 |
+
return processor.ocr_uploaded_image(image)
|
|
|
276 |
|
277 |
+
process_image_btn.click(
|
278 |
fn=process_image,
|
279 |
inputs=[processor_state, image_input],
|
280 |
outputs=[image_output, image_preview]
|
281 |
)
|
282 |
|
283 |
with gr.Tab("PDF OCR"):
|
284 |
+
with gr.Row():
|
285 |
+
pdf_input = gr.File(
|
286 |
+
label=f"Upload PDF (max {MAX_FILE_SIZE/1024/1024}MB, {MAX_PDF_PAGES} pages)",
|
287 |
+
file_types=SUPPORTED_PDF_TYPES
|
288 |
+
)
|
289 |
+
pdf_gallery = gr.Gallery(label="PDF Pages", height=300)
|
290 |
+
pdf_output = gr.Markdown(label="OCR Result", elem_classes="output-markdown")
|
291 |
+
process_pdf_btn = gr.Button("Process PDF", variant="primary")
|
292 |
|
293 |
def process_pdf(processor, pdf):
|
294 |
+
if not processor or not pdf:
|
295 |
+
return "Please set API key and upload a PDF", []
|
296 |
+
return processor.ocr_uploaded_pdf(pdf)
|
|
|
297 |
|
298 |
+
process_pdf_btn.click(
|
299 |
fn=process_pdf,
|
300 |
inputs=[processor_state, pdf_input],
|
301 |
outputs=[pdf_output, pdf_gallery]
|
302 |
)
|
303 |
|
304 |
+
with gr.Tab("Structured OCR"):
|
305 |
+
structured_input = gr.File(
|
306 |
+
label=f"Upload Image for Structured OCR (max {MAX_FILE_SIZE/1024/1024}MB)",
|
307 |
+
file_types=SUPPORTED_IMAGE_TYPES
|
308 |
+
)
|
309 |
+
structured_output = gr.Markdown(label="Structured Result", elem_classes="output-markdown")
|
310 |
+
structured_preview = gr.Image(label="Preview", height=300)
|
311 |
+
process_structured_btn = gr.Button("Process Structured OCR", variant="primary")
|
312 |
+
|
313 |
+
def process_structured(processor, image):
|
314 |
+
if not processor or not image:
|
315 |
+
return "Please set API key and upload an image", None
|
316 |
+
return processor.structured_ocr(image)
|
317 |
+
|
318 |
+
process_structured_btn.click(
|
319 |
+
fn=process_structured,
|
320 |
+
inputs=[processor_state, structured_input],
|
321 |
+
outputs=[structured_output, structured_preview]
|
322 |
+
)
|
323 |
|
324 |
+
return demo
|
325 |
|
326 |
if __name__ == "__main__":
|
327 |
+
os.environ['START_TIME'] = time.strftime('%Y-%m-%d %H:%M:%S')
|
328 |
+
print(f"===== Application Startup at {os.environ['START_TIME']} =====")
|
329 |
+
create_interface().launch(
|
330 |
+
share=True,
|
331 |
+
debug=True,
|
332 |
+
)
|