Spaces:
Running
on
Zero
Running
on
Zero
Commit
·
5639776
1
Parent(s):
2c499db
Refactor XML parsing functions for improved readability and consistency
Browse files
app.py
CHANGED
@@ -14,15 +14,14 @@ MODEL_LOAD_ERROR_MSG = None
|
|
14 |
|
15 |
HF_PROCESSOR = AutoProcessor.from_pretrained("reducto/RolmOCR")
|
16 |
HF_MODEL = AutoModelForImageTextToText.from_pretrained(
|
17 |
-
|
18 |
-
torch_dtype=torch.bfloat16,
|
19 |
-
device_map="auto"
|
20 |
)
|
21 |
HF_PIPE = pipeline("image-text-to-text", model=HF_MODEL, processor=HF_PROCESSOR)
|
22 |
|
23 |
|
24 |
# --- Helper Functions ---
|
25 |
|
|
|
26 |
def get_xml_namespace(xml_file_path):
|
27 |
"""
|
28 |
Dynamically gets the namespace from the XML file.
|
@@ -31,16 +30,17 @@ def get_xml_namespace(xml_file_path):
|
|
31 |
try:
|
32 |
tree = ET.parse(xml_file_path)
|
33 |
root = tree.getroot()
|
34 |
-
if
|
35 |
-
ns = root.tag.split(
|
36 |
# Determine format based on root element
|
37 |
-
if
|
38 |
-
return ns,
|
39 |
-
elif
|
40 |
-
return ns,
|
41 |
except ET.ParseError:
|
42 |
print(f"Error parsing XML to find namespace: {xml_file_path}")
|
43 |
-
return
|
|
|
44 |
|
45 |
def parse_page_xml_for_text(xml_file_path):
|
46 |
"""
|
@@ -49,7 +49,7 @@ def parse_page_xml_for_text(xml_file_path):
|
|
49 |
- full_text (str): All extracted text concatenated.
|
50 |
"""
|
51 |
full_text_lines = []
|
52 |
-
|
53 |
if not xml_file_path or not os.path.exists(xml_file_path):
|
54 |
return "Error: XML file not provided or does not exist."
|
55 |
|
@@ -59,23 +59,23 @@ def parse_page_xml_for_text(xml_file_path):
|
|
59 |
root = tree.getroot()
|
60 |
|
61 |
# Find all TextLine elements
|
62 |
-
for text_line in root.findall(f
|
63 |
# First try to get text from TextEquiv/Unicode
|
64 |
-
text_equiv = text_line.find(f
|
65 |
if text_equiv is not None and text_equiv.text:
|
66 |
full_text_lines.append(text_equiv.text)
|
67 |
continue
|
68 |
|
69 |
# If no TextEquiv, try to get text from Word elements
|
70 |
line_text_parts = []
|
71 |
-
for word in text_line.findall(f
|
72 |
-
word_text = word.find(f
|
73 |
if word_text is not None and word_text.text:
|
74 |
line_text_parts.append(word_text.text)
|
75 |
-
|
76 |
if line_text_parts:
|
77 |
full_text_lines.append(" ".join(line_text_parts))
|
78 |
-
|
79 |
return "\n".join(full_text_lines)
|
80 |
|
81 |
except ET.ParseError as e:
|
@@ -83,6 +83,7 @@ def parse_page_xml_for_text(xml_file_path):
|
|
83 |
except Exception as e:
|
84 |
return f"An unexpected error occurred during XML parsing: {e}"
|
85 |
|
|
|
86 |
def parse_alto_xml_for_text(xml_file_path):
|
87 |
"""
|
88 |
Parses an ALTO XML file to extract text content.
|
@@ -90,7 +91,7 @@ def parse_alto_xml_for_text(xml_file_path):
|
|
90 |
- full_text (str): All extracted text concatenated.
|
91 |
"""
|
92 |
full_text_lines = []
|
93 |
-
|
94 |
if not xml_file_path or not os.path.exists(xml_file_path):
|
95 |
return "Error: XML file not provided or does not exist."
|
96 |
|
@@ -99,15 +100,15 @@ def parse_alto_xml_for_text(xml_file_path):
|
|
99 |
tree = ET.parse(xml_file_path)
|
100 |
root = tree.getroot()
|
101 |
|
102 |
-
for text_line in root.findall(f
|
103 |
line_text_parts = []
|
104 |
-
for string_element in text_line.findall(f
|
105 |
-
text = string_element.get(
|
106 |
if text:
|
107 |
line_text_parts.append(text)
|
108 |
if line_text_parts:
|
109 |
full_text_lines.append(" ".join(line_text_parts))
|
110 |
-
|
111 |
return "\n".join(full_text_lines)
|
112 |
|
113 |
except ET.ParseError as e:
|
@@ -115,6 +116,7 @@ def parse_alto_xml_for_text(xml_file_path):
|
|
115 |
except Exception as e:
|
116 |
return f"An unexpected error occurred during XML parsing: {e}"
|
117 |
|
|
|
118 |
def parse_xml_for_text(xml_file_path):
|
119 |
"""
|
120 |
Main function to parse XML files, automatically detecting the format.
|
@@ -124,24 +126,29 @@ def parse_xml_for_text(xml_file_path):
|
|
124 |
|
125 |
try:
|
126 |
_, xml_format = get_xml_namespace(xml_file_path)
|
127 |
-
|
128 |
-
if xml_format ==
|
129 |
return parse_page_xml_for_text(xml_file_path)
|
130 |
-
elif xml_format ==
|
131 |
return parse_alto_xml_for_text(xml_file_path)
|
132 |
else:
|
133 |
return f"Error: Unsupported XML format. Expected ALTO or PAGE XML."
|
134 |
-
|
135 |
except Exception as e:
|
136 |
return f"Error determining XML format: {str(e)}"
|
137 |
|
|
|
138 |
@spaces.GPU
|
139 |
def predict(pil_image):
|
140 |
"""Performs OCR prediction using the Hugging Face model."""
|
141 |
global HF_PIPE, MODEL_LOAD_ERROR_MSG
|
142 |
|
143 |
if HF_PIPE is None:
|
144 |
-
error_to_report =
|
|
|
|
|
|
|
|
|
145 |
raise RuntimeError(error_to_report)
|
146 |
|
147 |
# Format the message in the expected structure
|
@@ -150,13 +157,17 @@ def predict(pil_image):
|
|
150 |
"role": "user",
|
151 |
"content": [
|
152 |
{"type": "image", "image": pil_image},
|
153 |
-
{
|
154 |
-
|
|
|
|
|
|
|
155 |
}
|
156 |
]
|
157 |
|
158 |
# Use the pipeline with the properly formatted messages
|
159 |
-
return HF_PIPE(messages,max_new_tokens=8096)
|
|
|
160 |
|
161 |
def run_hf_ocr(image_path):
|
162 |
"""
|
@@ -164,53 +175,68 @@ def run_hf_ocr(image_path):
|
|
164 |
"""
|
165 |
if image_path is None:
|
166 |
return "No image provided for OCR."
|
167 |
-
|
168 |
try:
|
169 |
pil_image = Image.open(image_path).convert("RGB")
|
170 |
-
ocr_results = predict(pil_image)
|
171 |
-
|
172 |
# Parse the output based on the user's example structure
|
173 |
-
if
|
174 |
-
|
175 |
-
|
|
|
|
|
|
|
|
|
176 |
if isinstance(generated_content, str):
|
177 |
return generated_content
|
178 |
|
179 |
if isinstance(generated_content, list) and generated_content:
|
180 |
if assistant_message := next(
|
181 |
(
|
182 |
-
msg[
|
183 |
for msg in reversed(generated_content)
|
184 |
if isinstance(msg, dict)
|
185 |
-
and msg.get(
|
186 |
-
and
|
187 |
),
|
188 |
None,
|
189 |
):
|
190 |
return assistant_message
|
191 |
-
|
192 |
# Fallback if the specific assistant message structure isn't found but there's content
|
193 |
-
if
|
194 |
-
|
195 |
-
|
196 |
-
|
197 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
198 |
|
199 |
print(f"Unexpected OCR output structure from HF model: {ocr_results}")
|
200 |
return "Error: Could not parse OCR model output. Check console."
|
201 |
-
|
202 |
else:
|
203 |
print(f"Unexpected OCR output structure from HF model: {ocr_results}")
|
204 |
return "Error: OCR model did not return expected output. Check console."
|
205 |
|
206 |
-
except RuntimeError as e:
|
207 |
return str(e)
|
208 |
except Exception as e:
|
209 |
print(f"Error during Hugging Face OCR processing: {e}")
|
210 |
return f"Error during Hugging Face OCR: {str(e)}"
|
211 |
|
|
|
212 |
# --- Gradio Interface Function ---
|
213 |
|
|
|
214 |
def process_files(image_path, xml_path):
|
215 |
"""
|
216 |
Main function for the Gradio interface.
|
@@ -226,7 +252,7 @@ def process_files(image_path, xml_path):
|
|
226 |
img_to_display = Image.open(image_path).convert("RGB")
|
227 |
hf_ocr_text_output = run_hf_ocr(image_path)
|
228 |
except Exception as e:
|
229 |
-
img_to_display = None
|
230 |
hf_ocr_text_output = f"Error loading image or running HF OCR: {e}"
|
231 |
else:
|
232 |
hf_ocr_text_output = "Please upload an image to perform OCR."
|
@@ -235,10 +261,10 @@ def process_files(image_path, xml_path):
|
|
235 |
xml_text_output = parse_xml_for_text(xml_path)
|
236 |
else:
|
237 |
xml_text_output = "No XML file uploaded."
|
238 |
-
|
239 |
# If only XML is provided without an image
|
240 |
if not image_path and xml_path:
|
241 |
-
img_to_display = None
|
242 |
hf_ocr_text_output = "Upload an image to perform OCR."
|
243 |
|
244 |
return img_to_display, xml_text_output, hf_ocr_text_output
|
@@ -255,38 +281,42 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
|
255 |
|
256 |
with gr.Row():
|
257 |
with gr.Column(scale=1):
|
258 |
-
image_input = gr.File(
|
259 |
-
|
|
|
|
|
|
|
|
|
260 |
submit_button = gr.Button("Process Image and XML", variant="primary")
|
261 |
|
262 |
with gr.Row():
|
263 |
with gr.Column(scale=1):
|
264 |
-
output_image_display = gr.Image(
|
|
|
|
|
265 |
with gr.Column(scale=1):
|
266 |
-
hf_ocr_output_textbox = gr.
|
267 |
-
label="OCR Output (Hugging Face Model)",
|
268 |
-
|
269 |
-
interactive=False,
|
270 |
-
show_copy_button=True
|
271 |
)
|
272 |
xml_output_textbox = gr.Textbox(
|
273 |
-
label="Text from XML",
|
274 |
-
lines=15,
|
275 |
interactive=False,
|
276 |
-
show_copy_button=True
|
277 |
)
|
278 |
-
|
279 |
submit_button.click(
|
280 |
fn=process_files,
|
281 |
inputs=[image_input, xml_input],
|
282 |
-
outputs=[output_image_display, xml_output_textbox, hf_ocr_output_textbox]
|
283 |
)
|
284 |
-
|
285 |
gr.Markdown("---")
|
286 |
gr.Markdown("### Example ALTO XML Snippet (for `String` element extraction):")
|
287 |
gr.Code(
|
288 |
value=(
|
289 |
-
"""<alto xmlns="http://www.loc.gov/standards/alto/v3/alto.xsd">
|
290 |
<Description>...</Description>
|
291 |
<Styles>...</Styles>
|
292 |
<Layout>
|
@@ -307,11 +337,13 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
|
307 |
</Layout>
|
308 |
</alto>"""
|
309 |
),
|
310 |
-
interactive=False
|
311 |
)
|
312 |
|
313 |
if __name__ == "__main__":
|
314 |
# Removed dummy file creation as it's less relevant for single file focus
|
315 |
print("Attempting to launch Gradio demo...")
|
316 |
-
print(
|
317 |
-
|
|
|
|
|
|
14 |
|
15 |
HF_PROCESSOR = AutoProcessor.from_pretrained("reducto/RolmOCR")
|
16 |
HF_MODEL = AutoModelForImageTextToText.from_pretrained(
|
17 |
+
"reducto/RolmOCR", torch_dtype=torch.bfloat16, device_map="auto"
|
|
|
|
|
18 |
)
|
19 |
HF_PIPE = pipeline("image-text-to-text", model=HF_MODEL, processor=HF_PROCESSOR)
|
20 |
|
21 |
|
22 |
# --- Helper Functions ---
|
23 |
|
24 |
+
|
25 |
def get_xml_namespace(xml_file_path):
|
26 |
"""
|
27 |
Dynamically gets the namespace from the XML file.
|
|
|
30 |
try:
|
31 |
tree = ET.parse(xml_file_path)
|
32 |
root = tree.getroot()
|
33 |
+
if "}" in root.tag:
|
34 |
+
ns = root.tag.split("}")[0] + "}"
|
35 |
# Determine format based on root element
|
36 |
+
if "PcGts" in root.tag:
|
37 |
+
return ns, "PAGE"
|
38 |
+
elif "alto" in root.tag.lower():
|
39 |
+
return ns, "ALTO"
|
40 |
except ET.ParseError:
|
41 |
print(f"Error parsing XML to find namespace: {xml_file_path}")
|
42 |
+
return "", "UNKNOWN"
|
43 |
+
|
44 |
|
45 |
def parse_page_xml_for_text(xml_file_path):
|
46 |
"""
|
|
|
49 |
- full_text (str): All extracted text concatenated.
|
50 |
"""
|
51 |
full_text_lines = []
|
52 |
+
|
53 |
if not xml_file_path or not os.path.exists(xml_file_path):
|
54 |
return "Error: XML file not provided or does not exist."
|
55 |
|
|
|
59 |
root = tree.getroot()
|
60 |
|
61 |
# Find all TextLine elements
|
62 |
+
for text_line in root.findall(f".//{ns_prefix}TextLine"):
|
63 |
# First try to get text from TextEquiv/Unicode
|
64 |
+
text_equiv = text_line.find(f"{ns_prefix}TextEquiv/{ns_prefix}Unicode")
|
65 |
if text_equiv is not None and text_equiv.text:
|
66 |
full_text_lines.append(text_equiv.text)
|
67 |
continue
|
68 |
|
69 |
# If no TextEquiv, try to get text from Word elements
|
70 |
line_text_parts = []
|
71 |
+
for word in text_line.findall(f"{ns_prefix}Word"):
|
72 |
+
word_text = word.find(f"{ns_prefix}TextEquiv/{ns_prefix}Unicode")
|
73 |
if word_text is not None and word_text.text:
|
74 |
line_text_parts.append(word_text.text)
|
75 |
+
|
76 |
if line_text_parts:
|
77 |
full_text_lines.append(" ".join(line_text_parts))
|
78 |
+
|
79 |
return "\n".join(full_text_lines)
|
80 |
|
81 |
except ET.ParseError as e:
|
|
|
83 |
except Exception as e:
|
84 |
return f"An unexpected error occurred during XML parsing: {e}"
|
85 |
|
86 |
+
|
87 |
def parse_alto_xml_for_text(xml_file_path):
|
88 |
"""
|
89 |
Parses an ALTO XML file to extract text content.
|
|
|
91 |
- full_text (str): All extracted text concatenated.
|
92 |
"""
|
93 |
full_text_lines = []
|
94 |
+
|
95 |
if not xml_file_path or not os.path.exists(xml_file_path):
|
96 |
return "Error: XML file not provided or does not exist."
|
97 |
|
|
|
100 |
tree = ET.parse(xml_file_path)
|
101 |
root = tree.getroot()
|
102 |
|
103 |
+
for text_line in root.findall(f".//{ns_prefix}TextLine"):
|
104 |
line_text_parts = []
|
105 |
+
for string_element in text_line.findall(f"{ns_prefix}String"):
|
106 |
+
text = string_element.get("CONTENT")
|
107 |
if text:
|
108 |
line_text_parts.append(text)
|
109 |
if line_text_parts:
|
110 |
full_text_lines.append(" ".join(line_text_parts))
|
111 |
+
|
112 |
return "\n".join(full_text_lines)
|
113 |
|
114 |
except ET.ParseError as e:
|
|
|
116 |
except Exception as e:
|
117 |
return f"An unexpected error occurred during XML parsing: {e}"
|
118 |
|
119 |
+
|
120 |
def parse_xml_for_text(xml_file_path):
|
121 |
"""
|
122 |
Main function to parse XML files, automatically detecting the format.
|
|
|
126 |
|
127 |
try:
|
128 |
_, xml_format = get_xml_namespace(xml_file_path)
|
129 |
+
|
130 |
+
if xml_format == "PAGE":
|
131 |
return parse_page_xml_for_text(xml_file_path)
|
132 |
+
elif xml_format == "ALTO":
|
133 |
return parse_alto_xml_for_text(xml_file_path)
|
134 |
else:
|
135 |
return f"Error: Unsupported XML format. Expected ALTO or PAGE XML."
|
136 |
+
|
137 |
except Exception as e:
|
138 |
return f"Error determining XML format: {str(e)}"
|
139 |
|
140 |
+
|
141 |
@spaces.GPU
|
142 |
def predict(pil_image):
|
143 |
"""Performs OCR prediction using the Hugging Face model."""
|
144 |
global HF_PIPE, MODEL_LOAD_ERROR_MSG
|
145 |
|
146 |
if HF_PIPE is None:
|
147 |
+
error_to_report = (
|
148 |
+
MODEL_LOAD_ERROR_MSG
|
149 |
+
if MODEL_LOAD_ERROR_MSG
|
150 |
+
else "OCR model could not be initialized."
|
151 |
+
)
|
152 |
raise RuntimeError(error_to_report)
|
153 |
|
154 |
# Format the message in the expected structure
|
|
|
157 |
"role": "user",
|
158 |
"content": [
|
159 |
{"type": "image", "image": pil_image},
|
160 |
+
{
|
161 |
+
"type": "text",
|
162 |
+
"text": "Return the plain text representation of this document as if you were reading it naturally.\n",
|
163 |
+
},
|
164 |
+
],
|
165 |
}
|
166 |
]
|
167 |
|
168 |
# Use the pipeline with the properly formatted messages
|
169 |
+
return HF_PIPE(messages, max_new_tokens=8096)
|
170 |
+
|
171 |
|
172 |
def run_hf_ocr(image_path):
|
173 |
"""
|
|
|
175 |
"""
|
176 |
if image_path is None:
|
177 |
return "No image provided for OCR."
|
178 |
+
|
179 |
try:
|
180 |
pil_image = Image.open(image_path).convert("RGB")
|
181 |
+
ocr_results = predict(pil_image) # predict handles model loading and inference
|
182 |
+
|
183 |
# Parse the output based on the user's example structure
|
184 |
+
if (
|
185 |
+
isinstance(ocr_results, list)
|
186 |
+
and ocr_results
|
187 |
+
and "generated_text" in ocr_results[0]
|
188 |
+
):
|
189 |
+
generated_content = ocr_results[0]["generated_text"]
|
190 |
+
|
191 |
if isinstance(generated_content, str):
|
192 |
return generated_content
|
193 |
|
194 |
if isinstance(generated_content, list) and generated_content:
|
195 |
if assistant_message := next(
|
196 |
(
|
197 |
+
msg["content"]
|
198 |
for msg in reversed(generated_content)
|
199 |
if isinstance(msg, dict)
|
200 |
+
and msg.get("role") == "assistant"
|
201 |
+
and "content" in msg
|
202 |
),
|
203 |
None,
|
204 |
):
|
205 |
return assistant_message
|
206 |
+
|
207 |
# Fallback if the specific assistant message structure isn't found but there's content
|
208 |
+
if (
|
209 |
+
isinstance(generated_content[0], dict)
|
210 |
+
and "content" in generated_content[0]
|
211 |
+
):
|
212 |
+
if (
|
213 |
+
len(generated_content) > 1
|
214 |
+
and isinstance(generated_content[1], dict)
|
215 |
+
and "content" in generated_content[1]
|
216 |
+
):
|
217 |
+
return generated_content[1][
|
218 |
+
"content"
|
219 |
+
] # Assuming second part is assistant
|
220 |
+
else:
|
221 |
+
return generated_content[0]["content"]
|
222 |
|
223 |
print(f"Unexpected OCR output structure from HF model: {ocr_results}")
|
224 |
return "Error: Could not parse OCR model output. Check console."
|
225 |
+
|
226 |
else:
|
227 |
print(f"Unexpected OCR output structure from HF model: {ocr_results}")
|
228 |
return "Error: OCR model did not return expected output. Check console."
|
229 |
|
230 |
+
except RuntimeError as e: # Catch model loading/initialization errors from predict
|
231 |
return str(e)
|
232 |
except Exception as e:
|
233 |
print(f"Error during Hugging Face OCR processing: {e}")
|
234 |
return f"Error during Hugging Face OCR: {str(e)}"
|
235 |
|
236 |
+
|
237 |
# --- Gradio Interface Function ---
|
238 |
|
239 |
+
|
240 |
def process_files(image_path, xml_path):
|
241 |
"""
|
242 |
Main function for the Gradio interface.
|
|
|
252 |
img_to_display = Image.open(image_path).convert("RGB")
|
253 |
hf_ocr_text_output = run_hf_ocr(image_path)
|
254 |
except Exception as e:
|
255 |
+
img_to_display = None # Clear image if it failed to load
|
256 |
hf_ocr_text_output = f"Error loading image or running HF OCR: {e}"
|
257 |
else:
|
258 |
hf_ocr_text_output = "Please upload an image to perform OCR."
|
|
|
261 |
xml_text_output = parse_xml_for_text(xml_path)
|
262 |
else:
|
263 |
xml_text_output = "No XML file uploaded."
|
264 |
+
|
265 |
# If only XML is provided without an image
|
266 |
if not image_path and xml_path:
|
267 |
+
img_to_display = None # No image to display
|
268 |
hf_ocr_text_output = "Upload an image to perform OCR."
|
269 |
|
270 |
return img_to_display, xml_text_output, hf_ocr_text_output
|
|
|
281 |
|
282 |
with gr.Row():
|
283 |
with gr.Column(scale=1):
|
284 |
+
image_input = gr.File(
|
285 |
+
label="Upload Image (PNG, JPG, etc.)", type="filepath"
|
286 |
+
)
|
287 |
+
xml_input = gr.File(
|
288 |
+
label="Upload XML File (Optional, ALTO or PAGE format)", type="filepath"
|
289 |
+
)
|
290 |
submit_button = gr.Button("Process Image and XML", variant="primary")
|
291 |
|
292 |
with gr.Row():
|
293 |
with gr.Column(scale=1):
|
294 |
+
output_image_display = gr.Image(
|
295 |
+
label="Uploaded Image", type="pil", interactive=False
|
296 |
+
)
|
297 |
with gr.Column(scale=1):
|
298 |
+
hf_ocr_output_textbox = gr.Markdown(
|
299 |
+
label="OCR Output (Hugging Face Model)",
|
300 |
+
show_copy_button=True,
|
|
|
|
|
301 |
)
|
302 |
xml_output_textbox = gr.Textbox(
|
303 |
+
label="Text from XML",
|
304 |
+
lines=15,
|
305 |
interactive=False,
|
306 |
+
show_copy_button=True,
|
307 |
)
|
308 |
+
|
309 |
submit_button.click(
|
310 |
fn=process_files,
|
311 |
inputs=[image_input, xml_input],
|
312 |
+
outputs=[output_image_display, xml_output_textbox, hf_ocr_output_textbox],
|
313 |
)
|
314 |
+
|
315 |
gr.Markdown("---")
|
316 |
gr.Markdown("### Example ALTO XML Snippet (for `String` element extraction):")
|
317 |
gr.Code(
|
318 |
value=(
|
319 |
+
"""<alto xmlns="http://www.loc.gov/standards/alto/v3/alto.xsd">
|
320 |
<Description>...</Description>
|
321 |
<Styles>...</Styles>
|
322 |
<Layout>
|
|
|
337 |
</Layout>
|
338 |
</alto>"""
|
339 |
),
|
340 |
+
interactive=False,
|
341 |
)
|
342 |
|
343 |
if __name__ == "__main__":
|
344 |
# Removed dummy file creation as it's less relevant for single file focus
|
345 |
print("Attempting to launch Gradio demo...")
|
346 |
+
print(
|
347 |
+
"If the Hugging Face model is large, initial startup might take some time due to model download/loading (on first OCR attempt)."
|
348 |
+
)
|
349 |
+
demo.launch()
|