Update app.py
Browse files
app.py
CHANGED
@@ -1,5 +1,4 @@
|
|
1 |
import gradio as gr
|
2 |
-
import yaml
|
3 |
import json
|
4 |
import base64
|
5 |
import tempfile
|
@@ -9,7 +8,8 @@ from datetime import datetime
|
|
9 |
from PIL import Image, ImageDraw, ImageFont
|
10 |
import io
|
11 |
import spaces
|
12 |
-
|
|
|
13 |
from htrflow.volume.volume import Collection
|
14 |
from htrflow.pipeline.pipeline import Pipeline
|
15 |
|
@@ -20,9 +20,7 @@ PIPELINE_CONFIGS = {
|
|
20 |
"step": "Segmentation",
|
21 |
"settings": {
|
22 |
"model": "yolo",
|
23 |
-
"model_settings": {
|
24 |
-
"model": "Riksarkivet/yolov9-lines-within-regions-1"
|
25 |
-
},
|
26 |
"generation_settings": {"batch_size": 8},
|
27 |
},
|
28 |
},
|
@@ -43,9 +41,7 @@ PIPELINE_CONFIGS = {
|
|
43 |
"step": "Segmentation",
|
44 |
"settings": {
|
45 |
"model": "yolo",
|
46 |
-
"model_settings": {
|
47 |
-
"model": "Riksarkivet/yolov9-lines-within-regions-1"
|
48 |
-
},
|
49 |
"generation_settings": {"batch_size": 8},
|
50 |
},
|
51 |
},
|
@@ -53,9 +49,7 @@ PIPELINE_CONFIGS = {
|
|
53 |
"step": "TextRecognition",
|
54 |
"settings": {
|
55 |
"model": "TrOCR",
|
56 |
-
"model_settings": {
|
57 |
-
"model": "Riksarkivet/trocr-base-handwritten-hist-swe-2"
|
58 |
-
},
|
59 |
"generation_settings": {"batch_size": 16},
|
60 |
},
|
61 |
},
|
@@ -76,9 +70,7 @@ PIPELINE_CONFIGS = {
|
|
76 |
"step": "Segmentation",
|
77 |
"settings": {
|
78 |
"model": "yolo",
|
79 |
-
"model_settings": {
|
80 |
-
"model": "Riksarkivet/yolov9-lines-within-regions-1"
|
81 |
-
},
|
82 |
"generation_settings": {"batch_size": 8},
|
83 |
},
|
84 |
},
|
@@ -107,9 +99,7 @@ PIPELINE_CONFIGS = {
|
|
107 |
"step": "Segmentation",
|
108 |
"settings": {
|
109 |
"model": "yolo",
|
110 |
-
"model_settings": {
|
111 |
-
"model": "Riksarkivet/yolov9-lines-within-regions-1"
|
112 |
-
},
|
113 |
"generation_settings": {"batch_size": 8},
|
114 |
},
|
115 |
},
|
@@ -117,9 +107,7 @@ PIPELINE_CONFIGS = {
|
|
117 |
"step": "TextRecognition",
|
118 |
"settings": {
|
119 |
"model": "TrOCR",
|
120 |
-
"model_settings": {
|
121 |
-
"model": "Riksarkivet/trocr-base-handwritten-hist-swe-2"
|
122 |
-
},
|
123 |
"generation_settings": {"batch_size": 16},
|
124 |
},
|
125 |
},
|
@@ -129,30 +117,8 @@ PIPELINE_CONFIGS = {
|
|
129 |
}
|
130 |
|
131 |
@spaces.GPU
|
132 |
-
def process_htr(
|
133 |
-
|
134 |
-
document_type: Literal[
|
135 |
-
"letter_english", "letter_swedish", "spread_english", "spread_swedish"
|
136 |
-
] = "spread_swedish",
|
137 |
-
confidence_threshold: float = 0.8,
|
138 |
-
custom_settings: Optional[str] = None,
|
139 |
-
) -> Dict:
|
140 |
-
"""
|
141 |
-
Process handwritten text recognition on uploaded images using HTRflow pipelines.
|
142 |
-
|
143 |
-
Supports templates for different document types (letters vs spreads) and
|
144 |
-
languages (English vs Swedish). Uses HTRflow's modular pipeline system with
|
145 |
-
configurable segmentation and text recognition models.
|
146 |
-
|
147 |
-
Args:
|
148 |
-
image (Image.Image): PIL Image object to process
|
149 |
-
document_type (str): Type of document processing template to use
|
150 |
-
confidence_threshold (float): Minimum confidence threshold for text recognition
|
151 |
-
custom_settings (str, optional): JSON string with custom pipeline settings
|
152 |
-
|
153 |
-
Returns:
|
154 |
-
dict: Processing results including extracted text, metadata, and processing state
|
155 |
-
"""
|
156 |
try:
|
157 |
if image is None:
|
158 |
return {"success": False, "error": "No image provided", "results": None}
|
@@ -166,30 +132,22 @@ def process_htr(
|
|
166 |
try:
|
167 |
config = json.loads(custom_settings)
|
168 |
except json.JSONDecodeError:
|
169 |
-
return {
|
170 |
-
"success": False,
|
171 |
-
"error": "Invalid JSON in custom_settings parameter",
|
172 |
-
"results": None,
|
173 |
-
}
|
174 |
else:
|
175 |
config = PIPELINE_CONFIGS[document_type]
|
176 |
|
177 |
collection = Collection([temp_image_path])
|
178 |
-
|
179 |
pipeline = Pipeline.from_config(config)
|
180 |
processed_collection = pipeline.run(collection)
|
181 |
|
182 |
-
results = extract_processing_results(
|
183 |
-
processed_collection, confidence_threshold
|
184 |
-
)
|
185 |
-
|
186 |
img_buffer = io.BytesIO()
|
187 |
image.save(img_buffer, format="PNG")
|
188 |
image_base64 = base64.b64encode(img_buffer.getvalue()).decode("utf-8")
|
189 |
|
|
|
|
|
190 |
processing_state = {
|
191 |
-
"
|
192 |
-
"config": config,
|
193 |
"image_base64": image_base64,
|
194 |
"image_size": image.size,
|
195 |
"document_type": document_type,
|
@@ -203,54 +161,22 @@ def process_htr(
|
|
203 |
"processing_state": json.dumps(processing_state),
|
204 |
"metadata": {
|
205 |
"total_lines": len(results.get("text_lines", [])),
|
206 |
-
"average_confidence":
|
207 |
"document_type": document_type,
|
208 |
"image_dimensions": image.size,
|
209 |
},
|
210 |
}
|
211 |
-
|
212 |
finally:
|
213 |
if os.path.exists(temp_image_path):
|
214 |
os.unlink(temp_image_path)
|
215 |
-
|
216 |
except Exception as e:
|
217 |
-
return {
|
218 |
-
"success": False,
|
219 |
-
"error": f"HTR processing failed: {str(e)}",
|
220 |
-
"results": None,
|
221 |
-
}
|
222 |
|
223 |
-
|
224 |
-
|
225 |
-
processing_state: str,
|
226 |
-
visualization_type: Literal[
|
227 |
-
"overlay", "confidence_heatmap", "text_regions"
|
228 |
-
] = "overlay",
|
229 |
-
show_confidence: bool = True,
|
230 |
-
highlight_low_confidence: bool = True,
|
231 |
-
image: Optional[Image.Image] = None,
|
232 |
-
) -> Dict:
|
233 |
-
"""
|
234 |
-
Generate interactive visualizations of HTR processing results.
|
235 |
-
|
236 |
-
Creates visual representations of text recognition results including bounding box
|
237 |
-
overlays, confidence heatmaps, and region segmentation displays. Supports multiple
|
238 |
-
visualization modes for different analysis needs.
|
239 |
-
|
240 |
-
Args:
|
241 |
-
processing_state (str): JSON string containing HTR processing results and metadata
|
242 |
-
visualization_type (str): Type of visualization to generate
|
243 |
-
show_confidence (bool): Whether to display confidence scores on visualization
|
244 |
-
highlight_low_confidence (bool): Whether to highlight low-confidence regions
|
245 |
-
image (Image.Image, optional): PIL Image object to use instead of state image
|
246 |
-
|
247 |
-
Returns:
|
248 |
-
dict: Visualization data including base64-encoded images and metadata
|
249 |
-
"""
|
250 |
try:
|
251 |
state = json.loads(processing_state)
|
252 |
-
|
253 |
-
confidence_threshold = state["confidence_threshold"]
|
254 |
|
255 |
if image is not None:
|
256 |
original_image = image
|
@@ -258,23 +184,12 @@ def visualize_results(
|
|
258 |
image_data = base64.b64decode(state["image_base64"])
|
259 |
original_image = Image.open(io.BytesIO(image_data))
|
260 |
|
261 |
-
|
262 |
-
viz_image = create_text_overlay_visualization(
|
263 |
-
original_image, collection, show_confidence, highlight_low_confidence
|
264 |
-
)
|
265 |
-
elif visualization_type == "confidence_heatmap":
|
266 |
-
viz_image = create_confidence_heatmap(
|
267 |
-
original_image, collection, confidence_threshold
|
268 |
-
)
|
269 |
-
elif visualization_type == "text_regions":
|
270 |
-
viz_image = create_region_visualization(original_image, collection)
|
271 |
|
272 |
img_buffer = io.BytesIO()
|
273 |
viz_image.save(img_buffer, format="PNG")
|
274 |
img_base64 = base64.b64encode(img_buffer.getvalue()).decode("utf-8")
|
275 |
|
276 |
-
viz_metadata = generate_visualization_metadata(collection, visualization_type)
|
277 |
-
|
278 |
return {
|
279 |
"success": True,
|
280 |
"visualization": {
|
@@ -283,521 +198,139 @@ def visualize_results(
|
|
283 |
"visualization_type": visualization_type,
|
284 |
"dimensions": viz_image.size,
|
285 |
},
|
286 |
-
"metadata":
|
287 |
-
"interactive_elements": extract_interactive_elements(collection),
|
288 |
}
|
289 |
-
|
290 |
except Exception as e:
|
291 |
-
return {
|
292 |
-
"success": False,
|
293 |
-
"error": f"Visualization generation failed: {str(e)}",
|
294 |
-
"visualization": None,
|
295 |
-
}
|
296 |
-
|
297 |
-
|
298 |
-
def export_results(
|
299 |
-
processing_state: str,
|
300 |
-
output_formats: List[Literal["txt", "json", "alto", "page"]] = ["txt"],
|
301 |
-
include_metadata: bool = True,
|
302 |
-
confidence_filter: float = 0.0,
|
303 |
-
) -> Dict:
|
304 |
-
"""
|
305 |
-
Export HTR results to multiple formats including plain text, structured JSON, ALTO XML, and PAGE XML.
|
306 |
|
307 |
-
|
308 |
-
|
309 |
-
|
310 |
-
Args:
|
311 |
-
processing_state (str): JSON string containing HTR processing results
|
312 |
-
output_formats (List[str]): List of output formats to generate
|
313 |
-
include_metadata (bool): Whether to include processing metadata in exports
|
314 |
-
confidence_filter (float): Minimum confidence threshold for included text
|
315 |
-
|
316 |
-
Returns:
|
317 |
-
dict: Export results with content for each requested format
|
318 |
-
"""
|
319 |
try:
|
320 |
-
# Parse processing state
|
321 |
state = json.loads(processing_state)
|
322 |
-
|
323 |
-
|
324 |
-
|
325 |
-
|
326 |
-
|
327 |
-
|
328 |
-
for format_type in output_formats:
|
329 |
-
if format_type == "txt":
|
330 |
-
exports["txt"] = export_plain_text(
|
331 |
-
collection, confidence_filter, include_metadata
|
332 |
-
)
|
333 |
-
elif format_type == "json":
|
334 |
-
exports["json"] = export_structured_json(
|
335 |
-
collection, confidence_filter, include_metadata
|
336 |
-
)
|
337 |
-
elif format_type == "alto":
|
338 |
-
exports["alto"] = export_alto_xml(
|
339 |
-
collection, confidence_filter, include_metadata
|
340 |
-
)
|
341 |
-
elif format_type == "page":
|
342 |
-
exports["page"] = export_page_xml(
|
343 |
-
collection, confidence_filter, include_metadata
|
344 |
-
)
|
345 |
|
346 |
-
|
347 |
-
|
|
|
|
|
348 |
|
349 |
-
|
350 |
-
|
351 |
-
|
352 |
-
|
353 |
-
|
354 |
-
|
355 |
-
|
356 |
-
|
357 |
-
|
358 |
-
|
359 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
360 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
361 |
except Exception as e:
|
362 |
-
return {
|
363 |
-
"success": False,
|
364 |
-
"error": f"Export generation failed: {str(e)}",
|
365 |
-
"exports": None,
|
366 |
-
}
|
367 |
-
|
368 |
-
|
369 |
-
# Helper Functions
|
370 |
-
def extract_processing_results(
|
371 |
-
collection: Collection, confidence_threshold: float
|
372 |
-
) -> Dict:
|
373 |
-
"""Extract structured results from processed HTRflow Collection."""
|
374 |
-
results = {
|
375 |
-
"extracted_text": "",
|
376 |
-
"text_lines": [],
|
377 |
-
"regions": [],
|
378 |
-
"confidence_scores": [],
|
379 |
-
}
|
380 |
|
381 |
-
|
|
|
382 |
for page in collection.pages:
|
383 |
for node in page.traverse():
|
384 |
-
if hasattr(node, "text") and node.text:
|
385 |
-
|
386 |
-
|
387 |
-
|
388 |
-
|
389 |
-
|
390 |
-
|
391 |
-
|
392 |
-
|
393 |
-
|
394 |
-
"node_id": getattr(node, "id", None),
|
395 |
-
}
|
396 |
-
)
|
397 |
-
results["extracted_text"] += node.text + "\n"
|
398 |
-
results["confidence_scores"].append(node.confidence)
|
399 |
-
|
400 |
return results
|
401 |
|
402 |
-
|
403 |
-
|
404 |
-
"""Serialize HTRflow Collection to JSON string for state storage."""
|
405 |
-
serialized_data = {"pages": [], "metadata": getattr(collection, "metadata", {})}
|
406 |
-
|
407 |
for page in collection.pages:
|
408 |
-
page_data = {
|
409 |
-
"nodes": [],
|
410 |
-
"image_path": getattr(page, "image_path", None),
|
411 |
-
"dimensions": getattr(page, "dimensions", None),
|
412 |
-
}
|
413 |
-
|
414 |
for node in page.traverse():
|
415 |
-
|
416 |
-
|
417 |
-
|
418 |
-
|
419 |
-
|
420 |
-
|
421 |
-
|
422 |
-
|
423 |
-
|
424 |
-
serialized_data["pages"].append(page_data)
|
425 |
-
|
426 |
-
return json.dumps(serialized_data)
|
427 |
-
|
428 |
-
|
429 |
-
def deserialize_collection(serialized_data: str):
|
430 |
-
"""Deserialize JSON string back to HTRflow Collection."""
|
431 |
-
data = json.loads(serialized_data)
|
432 |
-
|
433 |
-
# Mock collection classes for state reconstruction
|
434 |
-
class MockCollection:
|
435 |
-
def __init__(self, data):
|
436 |
-
self.pages = []
|
437 |
-
for page_data in data.get("pages", []):
|
438 |
-
page = MockPage(page_data)
|
439 |
-
self.pages.append(page)
|
440 |
-
|
441 |
-
class MockPage:
|
442 |
-
def __init__(self, page_data):
|
443 |
-
self.nodes = []
|
444 |
-
for node_data in page_data.get("nodes", []):
|
445 |
-
node = MockNode(node_data)
|
446 |
-
self.nodes.append(node)
|
447 |
-
|
448 |
-
def traverse(self):
|
449 |
-
return self.nodes
|
450 |
-
|
451 |
-
class MockNode:
|
452 |
-
def __init__(self, node_data):
|
453 |
-
self.text = node_data.get("text", "")
|
454 |
-
self.confidence = node_data.get("confidence", 1.0)
|
455 |
-
self.bbox = node_data.get("bbox")
|
456 |
-
self.id = node_data.get("node_id")
|
457 |
-
|
458 |
-
return MockCollection(data)
|
459 |
-
|
460 |
-
|
461 |
-
def calculate_average_confidence(results: Dict) -> float:
|
462 |
-
"""Calculate average confidence score from processing results."""
|
463 |
-
confidence_scores = results.get("confidence_scores", [])
|
464 |
-
if not confidence_scores:
|
465 |
-
return 0.0
|
466 |
-
return sum(confidence_scores) / len(confidence_scores)
|
467 |
-
|
468 |
-
|
469 |
-
def create_text_overlay_visualization(
|
470 |
-
image, collection, show_confidence, highlight_low_confidence
|
471 |
-
):
|
472 |
-
"""Create image with text bounding boxes and recognition results overlaid."""
|
473 |
viz_image = image.copy()
|
474 |
draw = ImageDraw.Draw(viz_image)
|
475 |
-
|
476 |
-
# Define visualization styles
|
477 |
-
bbox_color = (0, 255, 0) # Green for normal confidence
|
478 |
-
low_conf_color = (255, 165, 0) # Orange for low confidence
|
479 |
-
text_color = (255, 255, 255) # White text
|
480 |
-
|
481 |
try:
|
482 |
font = ImageFont.truetype("arial.ttf", 12)
|
483 |
except:
|
484 |
font = ImageFont.load_default()
|
485 |
|
486 |
-
|
487 |
-
|
488 |
-
|
489 |
-
|
490 |
-
|
491 |
-
|
492 |
-
and
|
493 |
-
and node.text
|
494 |
-
):
|
495 |
-
bbox = node.bbox
|
496 |
-
confidence = getattr(node, "confidence", 1.0)
|
497 |
-
|
498 |
-
# Choose color based on confidence
|
499 |
-
if highlight_low_confidence and confidence < 0.7:
|
500 |
-
color = low_conf_color
|
501 |
-
else:
|
502 |
-
color = bbox_color
|
503 |
-
|
504 |
-
# Draw bounding box
|
505 |
draw.rectangle(bbox, outline=color, width=2)
|
506 |
-
|
507 |
-
# Add confidence score if requested
|
508 |
if show_confidence:
|
509 |
-
|
510 |
-
|
511 |
-
|
512 |
-
return viz_image
|
513 |
-
|
514 |
-
|
515 |
-
def create_confidence_heatmap(image, collection, confidence_threshold):
|
516 |
-
"""Create confidence heatmap visualization."""
|
517 |
-
viz_image = image.copy()
|
518 |
-
|
519 |
-
# Create heatmap overlay based on confidence scores
|
520 |
-
for page in collection.pages:
|
521 |
-
for node in page.traverse():
|
522 |
-
if hasattr(node, "bbox") and hasattr(node, "confidence") and node.bbox:
|
523 |
-
confidence = node.confidence
|
524 |
-
# Color mapping: red (low) -> yellow (medium) -> green (high)
|
525 |
if confidence < 0.5:
|
526 |
-
color = (255, 0, 0, 100)
|
527 |
elif confidence < 0.8:
|
528 |
-
color = (255, 255, 0, 100)
|
529 |
else:
|
530 |
-
color = (0, 255, 0, 100)
|
531 |
-
|
532 |
-
# Create overlay image for transparency
|
533 |
overlay = Image.new("RGBA", viz_image.size, (0, 0, 0, 0))
|
534 |
overlay_draw = ImageDraw.Draw(overlay)
|
535 |
-
overlay_draw.rectangle(
|
536 |
viz_image = Image.alpha_composite(viz_image.convert("RGBA"), overlay)
|
|
|
|
|
|
|
|
|
|
|
537 |
|
538 |
-
return viz_image.convert("RGB")
|
539 |
-
|
540 |
-
|
541 |
-
def create_region_visualization(image, collection):
|
542 |
-
"""Create region segmentation visualization."""
|
543 |
-
viz_image = image.copy()
|
544 |
-
draw = ImageDraw.Draw(viz_image)
|
545 |
-
|
546 |
-
# Draw different colors for different region types
|
547 |
-
region_colors = [(255, 0, 0), (0, 255, 0), (0, 0, 255), (255, 255, 0)]
|
548 |
-
region_count = 0
|
549 |
-
|
550 |
-
for page in collection.pages:
|
551 |
-
for node in page.traverse():
|
552 |
-
if hasattr(node, "bbox") and node.bbox:
|
553 |
-
color = region_colors[region_count % len(region_colors)]
|
554 |
-
draw.rectangle(node.bbox, outline=color, width=3)
|
555 |
-
region_count += 1
|
556 |
-
|
557 |
-
return viz_image
|
558 |
-
|
559 |
-
|
560 |
-
def generate_visualization_metadata(collection, visualization_type):
|
561 |
-
"""Generate metadata for visualization results."""
|
562 |
-
total_elements = 0
|
563 |
-
confidence_stats = []
|
564 |
-
|
565 |
-
for page in collection.pages:
|
566 |
-
for node in page.traverse():
|
567 |
-
if hasattr(node, "text") and node.text:
|
568 |
-
total_elements += 1
|
569 |
-
if hasattr(node, "confidence"):
|
570 |
-
confidence_stats.append(node.confidence)
|
571 |
-
|
572 |
-
return {
|
573 |
-
"total_elements": total_elements,
|
574 |
-
"visualization_type": visualization_type,
|
575 |
-
"confidence_stats": {
|
576 |
-
"min": min(confidence_stats) if confidence_stats else 0,
|
577 |
-
"max": max(confidence_stats) if confidence_stats else 0,
|
578 |
-
"avg": sum(confidence_stats) / len(confidence_stats)
|
579 |
-
if confidence_stats
|
580 |
-
else 0,
|
581 |
-
},
|
582 |
-
}
|
583 |
-
|
584 |
-
|
585 |
-
def extract_interactive_elements(collection):
|
586 |
-
"""Extract interactive elements for visualization."""
|
587 |
-
elements = []
|
588 |
-
|
589 |
-
for page in collection.pages:
|
590 |
-
for node in page.traverse():
|
591 |
-
if (
|
592 |
-
hasattr(node, "bbox")
|
593 |
-
and hasattr(node, "text")
|
594 |
-
and node.bbox
|
595 |
-
and node.text
|
596 |
-
):
|
597 |
-
elements.append(
|
598 |
-
{
|
599 |
-
"bbox": node.bbox,
|
600 |
-
"text": node.text,
|
601 |
-
"confidence": getattr(node, "confidence", 1.0),
|
602 |
-
"node_id": getattr(node, "id", None),
|
603 |
-
}
|
604 |
-
)
|
605 |
-
|
606 |
-
return elements
|
607 |
-
|
608 |
-
|
609 |
-
def export_plain_text(
|
610 |
-
collection, confidence_filter: float, include_metadata: bool
|
611 |
-
) -> str:
|
612 |
-
"""Export recognition results as plain text."""
|
613 |
-
text_lines = []
|
614 |
-
|
615 |
-
if include_metadata:
|
616 |
-
text_lines.append(f"# HTR Export Results")
|
617 |
-
text_lines.append(f"# Confidence Filter: {confidence_filter}")
|
618 |
-
text_lines.append(f"# Export Time: {datetime.now().isoformat()}")
|
619 |
-
text_lines.append("")
|
620 |
-
|
621 |
-
# Extract text from collection hierarchy
|
622 |
-
for page in collection.pages:
|
623 |
-
for node in page.traverse():
|
624 |
-
if hasattr(node, "text") and node.text:
|
625 |
-
confidence = getattr(node, "confidence", 1.0)
|
626 |
-
if confidence >= confidence_filter:
|
627 |
-
text_lines.append(node.text)
|
628 |
-
|
629 |
-
return "\n".join(text_lines)
|
630 |
-
|
631 |
|
632 |
-
def export_structured_json(
|
633 |
-
collection, confidence_filter: float, include_metadata: bool
|
634 |
-
) -> str:
|
635 |
-
"""Export results as structured JSON with full hierarchy."""
|
636 |
-
result = {"document": {"pages": []}}
|
637 |
-
|
638 |
-
if include_metadata:
|
639 |
-
result["metadata"] = {
|
640 |
-
"confidence_filter": confidence_filter,
|
641 |
-
"export_time": datetime.now().isoformat(),
|
642 |
-
"total_pages": len(collection.pages),
|
643 |
-
}
|
644 |
-
|
645 |
-
# Build hierarchical structure
|
646 |
-
for page_idx, page in enumerate(collection.pages):
|
647 |
-
page_data = {"page_id": page_idx, "regions": []}
|
648 |
-
|
649 |
-
for node in page.traverse():
|
650 |
-
if hasattr(node, "text") and node.text:
|
651 |
-
confidence = getattr(node, "confidence", 1.0)
|
652 |
-
if confidence >= confidence_filter:
|
653 |
-
node_data = {
|
654 |
-
"text": node.text,
|
655 |
-
"confidence": confidence,
|
656 |
-
"bbox": getattr(node, "bbox", None),
|
657 |
-
"node_id": getattr(node, "id", None),
|
658 |
-
}
|
659 |
-
page_data["regions"].append(node_data)
|
660 |
-
|
661 |
-
result["document"]["pages"].append(page_data)
|
662 |
-
|
663 |
-
return json.dumps(result, indent=2, ensure_ascii=False)
|
664 |
-
|
665 |
-
|
666 |
-
def export_alto_xml(
|
667 |
-
collection, confidence_filter: float, include_metadata: bool
|
668 |
-
) -> str:
|
669 |
-
"""Export results as ALTO XML format."""
|
670 |
-
# Simplified ALTO XML generation
|
671 |
-
xml_lines = ['<?xml version="1.0" encoding="UTF-8"?>']
|
672 |
-
xml_lines.append('<alto xmlns="http://www.loc.gov/standards/alto/ns-v4#">')
|
673 |
-
xml_lines.append(" <Description>")
|
674 |
-
if include_metadata:
|
675 |
-
xml_lines.append(f" <sourceImageInformation>")
|
676 |
-
xml_lines.append(f" <fileName>htr_processed_image</fileName>")
|
677 |
-
xml_lines.append(f" </sourceImageInformation>")
|
678 |
-
xml_lines.append(" </Description>")
|
679 |
-
xml_lines.append(" <Layout>")
|
680 |
-
xml_lines.append(" <Page>")
|
681 |
-
|
682 |
-
for page in collection.pages:
|
683 |
-
for node in page.traverse():
|
684 |
-
if hasattr(node, "text") and node.text:
|
685 |
-
confidence = getattr(node, "confidence", 1.0)
|
686 |
-
if confidence >= confidence_filter:
|
687 |
-
bbox = getattr(node, "bbox", [0, 0, 100, 20])
|
688 |
-
xml_lines.append(
|
689 |
-
f' <TextLine HPOS="{bbox[0]}" VPOS="{bbox[1]}" WIDTH="{bbox[2] - bbox[0]}" HEIGHT="{bbox[3] - bbox[1]}">'
|
690 |
-
)
|
691 |
-
xml_lines.append(
|
692 |
-
f' <String CONTENT="{node.text}" WC="{confidence:.3f}"/>'
|
693 |
-
)
|
694 |
-
xml_lines.append(" </TextLine>")
|
695 |
-
|
696 |
-
xml_lines.append(" </Page>")
|
697 |
-
xml_lines.append(" </Layout>")
|
698 |
-
xml_lines.append("</alto>")
|
699 |
-
|
700 |
-
return "\n".join(xml_lines)
|
701 |
-
|
702 |
-
|
703 |
-
def export_page_xml(
|
704 |
-
collection, confidence_filter: float, include_metadata: bool
|
705 |
-
) -> str:
|
706 |
-
"""Export results as PAGE XML format."""
|
707 |
-
# Simplified PAGE XML generation
|
708 |
-
xml_lines = ['<?xml version="1.0" encoding="UTF-8"?>']
|
709 |
-
xml_lines.append(
|
710 |
-
'<PcGts xmlns="http://schema.primaresearch.org/PAGE/gts/pagecontent/2013-07-15">'
|
711 |
-
)
|
712 |
-
if include_metadata:
|
713 |
-
xml_lines.append(" <Metadata>")
|
714 |
-
xml_lines.append(f" <Created>{datetime.now().isoformat()}</Created>")
|
715 |
-
xml_lines.append(" </Metadata>")
|
716 |
-
xml_lines.append(" <Page>")
|
717 |
-
|
718 |
-
for page in collection.pages:
|
719 |
-
for node in page.traverse():
|
720 |
-
if hasattr(node, "text") and node.text:
|
721 |
-
confidence = getattr(node, "confidence", 1.0)
|
722 |
-
if confidence >= confidence_filter:
|
723 |
-
bbox = getattr(node, "bbox", [0, 0, 100, 20])
|
724 |
-
xml_lines.append(f" <TextRegion>")
|
725 |
-
xml_lines.append(
|
726 |
-
f' <Coords points="{bbox[0]},{bbox[1]} {bbox[2]},{bbox[1]} {bbox[2]},{bbox[3]} {bbox[0]},{bbox[3]}"/>'
|
727 |
-
)
|
728 |
-
xml_lines.append(f" <TextLine>")
|
729 |
-
xml_lines.append(f' <TextEquiv conf="{confidence:.3f}">')
|
730 |
-
xml_lines.append(f" <Unicode>{node.text}</Unicode>")
|
731 |
-
xml_lines.append(" </TextEquiv>")
|
732 |
-
xml_lines.append(" </TextLine>")
|
733 |
-
xml_lines.append(" </TextRegion>")
|
734 |
-
|
735 |
-
xml_lines.append(" </Page>")
|
736 |
-
xml_lines.append("</PcGts>")
|
737 |
-
|
738 |
-
return "\n".join(xml_lines)
|
739 |
-
|
740 |
-
|
741 |
-
def calculate_export_statistics(collection, confidence_filter: float) -> Dict:
|
742 |
-
"""Calculate statistics for export results."""
|
743 |
-
total_text_elements = 0
|
744 |
-
filtered_text_elements = 0
|
745 |
-
confidence_scores = []
|
746 |
-
total_characters = 0
|
747 |
-
|
748 |
-
for page in collection.pages:
|
749 |
-
for node in page.traverse():
|
750 |
-
if hasattr(node, "text") and node.text:
|
751 |
-
total_text_elements += 1
|
752 |
-
confidence = getattr(node, "confidence", 1.0)
|
753 |
-
confidence_scores.append(confidence)
|
754 |
-
|
755 |
-
if confidence >= confidence_filter:
|
756 |
-
filtered_text_elements += 1
|
757 |
-
total_characters += len(node.text)
|
758 |
-
|
759 |
-
return {
|
760 |
-
"total_text_elements": total_text_elements,
|
761 |
-
"filtered_text_elements": filtered_text_elements,
|
762 |
-
"filter_retention_rate": filtered_text_elements / total_text_elements
|
763 |
-
if total_text_elements > 0
|
764 |
-
else 0,
|
765 |
-
"total_characters": total_characters,
|
766 |
-
"average_confidence": sum(confidence_scores) / len(confidence_scores)
|
767 |
-
if confidence_scores
|
768 |
-
else 0,
|
769 |
-
"confidence_range": {
|
770 |
-
"min": min(confidence_scores) if confidence_scores else 0,
|
771 |
-
"max": max(confidence_scores) if confidence_scores else 0,
|
772 |
-
},
|
773 |
-
}
|
774 |
-
|
775 |
-
|
776 |
-
# Main Gradio Application with MCP Server
|
777 |
def create_htrflow_mcp_server():
|
778 |
-
"""Create the complete HTRflow MCP server with all three tools."""
|
779 |
-
|
780 |
demo = gr.TabbedInterface(
|
781 |
[
|
782 |
gr.Interface(
|
783 |
fn=process_htr,
|
784 |
inputs=[
|
785 |
gr.Image(type="pil", label="Upload Image"),
|
786 |
-
gr.Dropdown(
|
787 |
-
choices=[
|
788 |
-
"letter_english",
|
789 |
-
"letter_swedish",
|
790 |
-
"spread_english",
|
791 |
-
"spread_swedish",
|
792 |
-
],
|
793 |
-
value="letter_english",
|
794 |
-
label="Document Type",
|
795 |
-
),
|
796 |
gr.Slider(0.0, 1.0, value=0.8, label="Confidence Threshold"),
|
797 |
-
gr.Textbox(
|
798 |
-
label="Custom Settings (JSON)",
|
799 |
-
placeholder="Optional custom pipeline settings",
|
800 |
-
),
|
801 |
],
|
802 |
outputs=gr.JSON(label="Processing Results"),
|
803 |
title="HTR Processing Tool",
|
@@ -807,21 +340,11 @@ def create_htrflow_mcp_server():
|
|
807 |
gr.Interface(
|
808 |
fn=visualize_results,
|
809 |
inputs=[
|
810 |
-
gr.Textbox(
|
811 |
-
|
812 |
-
placeholder="Paste processing results from HTR tool",
|
813 |
-
),
|
814 |
-
gr.Dropdown(
|
815 |
-
choices=["overlay", "confidence_heatmap", "text_regions"],
|
816 |
-
value="overlay",
|
817 |
-
label="Visualization Type",
|
818 |
-
),
|
819 |
gr.Checkbox(value=True, label="Show Confidence Scores"),
|
820 |
gr.Checkbox(value=True, label="Highlight Low Confidence"),
|
821 |
-
gr.Image(
|
822 |
-
type="pil",
|
823 |
-
label="Image (optional - will use image from processing state if not provided)",
|
824 |
-
),
|
825 |
],
|
826 |
outputs=gr.JSON(label="Visualization Results"),
|
827 |
title="Results Visualization Tool",
|
@@ -831,16 +354,8 @@ def create_htrflow_mcp_server():
|
|
831 |
gr.Interface(
|
832 |
fn=export_results,
|
833 |
inputs=[
|
834 |
-
gr.Textbox(
|
835 |
-
|
836 |
-
placeholder="Paste processing results from HTR tool",
|
837 |
-
),
|
838 |
-
gr.CheckboxGroup(
|
839 |
-
choices=["txt", "json", "alto", "page"],
|
840 |
-
value=["txt"],
|
841 |
-
label="Output Formats",
|
842 |
-
),
|
843 |
-
gr.Checkbox(value=True, label="Include Metadata"),
|
844 |
gr.Slider(0.0, 1.0, value=0.0, label="Confidence Filter"),
|
845 |
],
|
846 |
outputs=gr.JSON(label="Export Results"),
|
@@ -852,11 +367,8 @@ def create_htrflow_mcp_server():
|
|
852 |
["HTR Processing", "Results Visualization", "Export Results"],
|
853 |
title="HTRflow MCP Server",
|
854 |
)
|
855 |
-
|
856 |
return demo
|
857 |
|
858 |
-
|
859 |
-
# Launch MCP Server
|
860 |
if __name__ == "__main__":
|
861 |
demo = create_htrflow_mcp_server()
|
862 |
-
demo.launch(mcp_server=True)
|
|
|
1 |
import gradio as gr
|
|
|
2 |
import json
|
3 |
import base64
|
4 |
import tempfile
|
|
|
8 |
from PIL import Image, ImageDraw, ImageFont
|
9 |
import io
|
10 |
import spaces
|
11 |
+
import shutil
|
12 |
+
from pathlib import Path
|
13 |
from htrflow.volume.volume import Collection
|
14 |
from htrflow.pipeline.pipeline import Pipeline
|
15 |
|
|
|
20 |
"step": "Segmentation",
|
21 |
"settings": {
|
22 |
"model": "yolo",
|
23 |
+
"model_settings": {"model": "Riksarkivet/yolov9-lines-within-regions-1"},
|
|
|
|
|
24 |
"generation_settings": {"batch_size": 8},
|
25 |
},
|
26 |
},
|
|
|
41 |
"step": "Segmentation",
|
42 |
"settings": {
|
43 |
"model": "yolo",
|
44 |
+
"model_settings": {"model": "Riksarkivet/yolov9-lines-within-regions-1"},
|
|
|
|
|
45 |
"generation_settings": {"batch_size": 8},
|
46 |
},
|
47 |
},
|
|
|
49 |
"step": "TextRecognition",
|
50 |
"settings": {
|
51 |
"model": "TrOCR",
|
52 |
+
"model_settings": {"model": "Riksarkivet/trocr-base-handwritten-hist-swe-2"},
|
|
|
|
|
53 |
"generation_settings": {"batch_size": 16},
|
54 |
},
|
55 |
},
|
|
|
70 |
"step": "Segmentation",
|
71 |
"settings": {
|
72 |
"model": "yolo",
|
73 |
+
"model_settings": {"model": "Riksarkivet/yolov9-lines-within-regions-1"},
|
|
|
|
|
74 |
"generation_settings": {"batch_size": 8},
|
75 |
},
|
76 |
},
|
|
|
99 |
"step": "Segmentation",
|
100 |
"settings": {
|
101 |
"model": "yolo",
|
102 |
+
"model_settings": {"model": "Riksarkivet/yolov9-lines-within-regions-1"},
|
|
|
|
|
103 |
"generation_settings": {"batch_size": 8},
|
104 |
},
|
105 |
},
|
|
|
107 |
"step": "TextRecognition",
|
108 |
"settings": {
|
109 |
"model": "TrOCR",
|
110 |
+
"model_settings": {"model": "Riksarkivet/trocr-base-handwritten-hist-swe-2"},
|
|
|
|
|
111 |
"generation_settings": {"batch_size": 16},
|
112 |
},
|
113 |
},
|
|
|
117 |
}
|
118 |
|
119 |
@spaces.GPU
|
120 |
+
def process_htr(image: Image.Image, document_type: Literal["letter_english", "letter_swedish", "spread_english", "spread_swedish"] = "spread_swedish", confidence_threshold: float = 0.8, custom_settings: Optional[str] = None) -> Dict:
|
121 |
+
"""Process handwritten text recognition on uploaded images using HTRflow pipelines."""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
122 |
try:
|
123 |
if image is None:
|
124 |
return {"success": False, "error": "No image provided", "results": None}
|
|
|
132 |
try:
|
133 |
config = json.loads(custom_settings)
|
134 |
except json.JSONDecodeError:
|
135 |
+
return {"success": False, "error": "Invalid JSON in custom_settings parameter", "results": None}
|
|
|
|
|
|
|
|
|
136 |
else:
|
137 |
config = PIPELINE_CONFIGS[document_type]
|
138 |
|
139 |
collection = Collection([temp_image_path])
|
|
|
140 |
pipeline = Pipeline.from_config(config)
|
141 |
processed_collection = pipeline.run(collection)
|
142 |
|
|
|
|
|
|
|
|
|
143 |
img_buffer = io.BytesIO()
|
144 |
image.save(img_buffer, format="PNG")
|
145 |
image_base64 = base64.b64encode(img_buffer.getvalue()).decode("utf-8")
|
146 |
|
147 |
+
results = extract_text_results(processed_collection, confidence_threshold)
|
148 |
+
|
149 |
processing_state = {
|
150 |
+
"collection_data": serialize_collection_data(processed_collection),
|
|
|
151 |
"image_base64": image_base64,
|
152 |
"image_size": image.size,
|
153 |
"document_type": document_type,
|
|
|
161 |
"processing_state": json.dumps(processing_state),
|
162 |
"metadata": {
|
163 |
"total_lines": len(results.get("text_lines", [])),
|
164 |
+
"average_confidence": results.get("average_confidence", 0),
|
165 |
"document_type": document_type,
|
166 |
"image_dimensions": image.size,
|
167 |
},
|
168 |
}
|
|
|
169 |
finally:
|
170 |
if os.path.exists(temp_image_path):
|
171 |
os.unlink(temp_image_path)
|
|
|
172 |
except Exception as e:
|
173 |
+
return {"success": False, "error": f"HTR processing failed: {str(e)}", "results": None}
|
|
|
|
|
|
|
|
|
174 |
|
175 |
+
def visualize_results(processing_state: str, visualization_type: Literal["overlay", "confidence_heatmap", "text_regions"] = "overlay", show_confidence: bool = True, highlight_low_confidence: bool = True, image: Optional[Image.Image] = None) -> Dict:
|
176 |
+
"""Generate interactive visualizations of HTR processing results."""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
177 |
try:
|
178 |
state = json.loads(processing_state)
|
179 |
+
collection_data = state["collection_data"]
|
|
|
180 |
|
181 |
if image is not None:
|
182 |
original_image = image
|
|
|
184 |
image_data = base64.b64decode(state["image_base64"])
|
185 |
original_image = Image.open(io.BytesIO(image_data))
|
186 |
|
187 |
+
viz_image = create_visualization(original_image, collection_data, visualization_type, show_confidence, highlight_low_confidence)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
188 |
|
189 |
img_buffer = io.BytesIO()
|
190 |
viz_image.save(img_buffer, format="PNG")
|
191 |
img_base64 = base64.b64encode(img_buffer.getvalue()).decode("utf-8")
|
192 |
|
|
|
|
|
193 |
return {
|
194 |
"success": True,
|
195 |
"visualization": {
|
|
|
198 |
"visualization_type": visualization_type,
|
199 |
"dimensions": viz_image.size,
|
200 |
},
|
201 |
+
"metadata": {"total_elements": len(collection_data.get("text_elements", []))},
|
|
|
202 |
}
|
|
|
203 |
except Exception as e:
|
204 |
+
return {"success": False, "error": f"Visualization generation failed: {str(e)}", "visualization": None}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
205 |
|
206 |
+
def export_results(processing_state: str, output_formats: List[Literal["txt", "json", "alto", "page"]] = ["txt"], confidence_filter: float = 0.0) -> Dict:
|
207 |
+
"""Export HTR results to multiple formats using HTRflow's native export functionality."""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
208 |
try:
|
|
|
209 |
state = json.loads(processing_state)
|
210 |
+
|
211 |
+
with tempfile.NamedTemporaryFile(suffix=".png", delete=False) as temp_file:
|
212 |
+
image_data = base64.b64decode(state["image_base64"])
|
213 |
+
image = Image.open(io.BytesIO(image_data))
|
214 |
+
image.save(temp_file.name, "PNG")
|
215 |
+
temp_image_path = temp_file.name
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
216 |
|
217 |
+
try:
|
218 |
+
collection = Collection([temp_image_path])
|
219 |
+
pipeline = Pipeline.from_config(PIPELINE_CONFIGS[state["document_type"]])
|
220 |
+
processed_collection = pipeline.run(collection)
|
221 |
|
222 |
+
temp_dir = Path(tempfile.mkdtemp())
|
223 |
+
exports = {}
|
224 |
+
|
225 |
+
for fmt in output_formats:
|
226 |
+
export_dir = temp_dir / fmt
|
227 |
+
processed_collection.save(directory=str(export_dir), serializer=fmt)
|
228 |
+
|
229 |
+
export_files = []
|
230 |
+
for root, _, files in os.walk(export_dir):
|
231 |
+
for file in files:
|
232 |
+
file_path = os.path.join(root, file)
|
233 |
+
with open(file_path, 'r', encoding='utf-8') as f:
|
234 |
+
content = f.read()
|
235 |
+
export_files.append({"filename": file, "content": content})
|
236 |
+
|
237 |
+
exports[fmt] = export_files
|
238 |
+
|
239 |
+
shutil.rmtree(temp_dir)
|
240 |
|
241 |
+
return {
|
242 |
+
"success": True,
|
243 |
+
"exports": exports,
|
244 |
+
"export_metadata": {
|
245 |
+
"formats_generated": output_formats,
|
246 |
+
"confidence_filter": confidence_filter,
|
247 |
+
"timestamp": datetime.now().isoformat(),
|
248 |
+
},
|
249 |
+
}
|
250 |
+
finally:
|
251 |
+
if os.path.exists(temp_image_path):
|
252 |
+
os.unlink(temp_image_path)
|
253 |
+
|
254 |
except Exception as e:
|
255 |
+
return {"success": False, "error": f"Export generation failed: {str(e)}", "exports": None}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
256 |
|
257 |
+
def extract_text_results(collection: Collection, confidence_threshold: float) -> Dict:
|
258 |
+
results = {"extracted_text": "", "text_lines": [], "confidence_scores": []}
|
259 |
for page in collection.pages:
|
260 |
for node in page.traverse():
|
261 |
+
if hasattr(node, "text") and node.text and hasattr(node, "confidence") and node.confidence >= confidence_threshold:
|
262 |
+
results["text_lines"].append({
|
263 |
+
"text": node.text,
|
264 |
+
"confidence": node.confidence,
|
265 |
+
"bbox": getattr(node, "bbox", None),
|
266 |
+
})
|
267 |
+
results["extracted_text"] += node.text + "\n"
|
268 |
+
results["confidence_scores"].append(node.confidence)
|
269 |
+
|
270 |
+
results["average_confidence"] = sum(results["confidence_scores"]) / len(results["confidence_scores"]) if results["confidence_scores"] else 0
|
|
|
|
|
|
|
|
|
|
|
|
|
271 |
return results
|
272 |
|
273 |
+
def serialize_collection_data(collection: Collection) -> Dict:
|
274 |
+
text_elements = []
|
|
|
|
|
|
|
275 |
for page in collection.pages:
|
|
|
|
|
|
|
|
|
|
|
|
|
276 |
for node in page.traverse():
|
277 |
+
if hasattr(node, "text") and node.text:
|
278 |
+
text_elements.append({
|
279 |
+
"text": node.text,
|
280 |
+
"confidence": getattr(node, "confidence", 1.0),
|
281 |
+
"bbox": getattr(node, "bbox", None),
|
282 |
+
})
|
283 |
+
return {"text_elements": text_elements}
|
284 |
+
|
285 |
+
def create_visualization(image, collection_data, visualization_type, show_confidence, highlight_low_confidence):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
286 |
viz_image = image.copy()
|
287 |
draw = ImageDraw.Draw(viz_image)
|
288 |
+
|
|
|
|
|
|
|
|
|
|
|
289 |
try:
|
290 |
font = ImageFont.truetype("arial.ttf", 12)
|
291 |
except:
|
292 |
font = ImageFont.load_default()
|
293 |
|
294 |
+
for element in collection_data.get("text_elements", []):
|
295 |
+
if element.get("bbox"):
|
296 |
+
bbox = element["bbox"]
|
297 |
+
confidence = element.get("confidence", 1.0)
|
298 |
+
|
299 |
+
if visualization_type == "overlay":
|
300 |
+
color = (255, 165, 0) if highlight_low_confidence and confidence < 0.7 else (0, 255, 0)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
301 |
draw.rectangle(bbox, outline=color, width=2)
|
|
|
|
|
302 |
if show_confidence:
|
303 |
+
draw.text((bbox[0], bbox[1] - 15), f"{confidence:.2f}", fill=color, font=font)
|
304 |
+
|
305 |
+
elif visualization_type == "confidence_heatmap":
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
306 |
if confidence < 0.5:
|
307 |
+
color = (255, 0, 0, 100)
|
308 |
elif confidence < 0.8:
|
309 |
+
color = (255, 255, 0, 100)
|
310 |
else:
|
311 |
+
color = (0, 255, 0, 100)
|
|
|
|
|
312 |
overlay = Image.new("RGBA", viz_image.size, (0, 0, 0, 0))
|
313 |
overlay_draw = ImageDraw.Draw(overlay)
|
314 |
+
overlay_draw.rectangle(bbox, fill=color)
|
315 |
viz_image = Image.alpha_composite(viz_image.convert("RGBA"), overlay)
|
316 |
+
|
317 |
+
elif visualization_type == "text_regions":
|
318 |
+
colors = [(255, 0, 0), (0, 255, 0), (0, 0, 255), (255, 255, 0)]
|
319 |
+
color = colors[hash(str(bbox)) % len(colors)]
|
320 |
+
draw.rectangle(bbox, outline=color, width=3)
|
321 |
|
322 |
+
return viz_image.convert("RGB") if visualization_type == "confidence_heatmap" else viz_image
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
323 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
324 |
def create_htrflow_mcp_server():
|
|
|
|
|
325 |
demo = gr.TabbedInterface(
|
326 |
[
|
327 |
gr.Interface(
|
328 |
fn=process_htr,
|
329 |
inputs=[
|
330 |
gr.Image(type="pil", label="Upload Image"),
|
331 |
+
gr.Dropdown(choices=["letter_english", "letter_swedish", "spread_english", "spread_swedish"], value="letter_english", label="Document Type"),
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
332 |
gr.Slider(0.0, 1.0, value=0.8, label="Confidence Threshold"),
|
333 |
+
gr.Textbox(label="Custom Settings (JSON)", placeholder="Optional custom pipeline settings"),
|
|
|
|
|
|
|
334 |
],
|
335 |
outputs=gr.JSON(label="Processing Results"),
|
336 |
title="HTR Processing Tool",
|
|
|
340 |
gr.Interface(
|
341 |
fn=visualize_results,
|
342 |
inputs=[
|
343 |
+
gr.Textbox(label="Processing State (JSON)", placeholder="Paste processing results from HTR tool"),
|
344 |
+
gr.Dropdown(choices=["overlay", "confidence_heatmap", "text_regions"], value="overlay", label="Visualization Type"),
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
345 |
gr.Checkbox(value=True, label="Show Confidence Scores"),
|
346 |
gr.Checkbox(value=True, label="Highlight Low Confidence"),
|
347 |
+
gr.Image(type="pil", label="Image (optional)"),
|
|
|
|
|
|
|
348 |
],
|
349 |
outputs=gr.JSON(label="Visualization Results"),
|
350 |
title="Results Visualization Tool",
|
|
|
354 |
gr.Interface(
|
355 |
fn=export_results,
|
356 |
inputs=[
|
357 |
+
gr.Textbox(label="Processing State (JSON)", placeholder="Paste processing results from HTR tool"),
|
358 |
+
gr.CheckboxGroup(choices=["txt", "json", "alto", "page"], value=["txt"], label="Output Formats"),
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
359 |
gr.Slider(0.0, 1.0, value=0.0, label="Confidence Filter"),
|
360 |
],
|
361 |
outputs=gr.JSON(label="Export Results"),
|
|
|
367 |
["HTR Processing", "Results Visualization", "Export Results"],
|
368 |
title="HTRflow MCP Server",
|
369 |
)
|
|
|
370 |
return demo
|
371 |
|
|
|
|
|
372 |
if __name__ == "__main__":
|
373 |
demo = create_htrflow_mcp_server()
|
374 |
+
demo.launch(mcp_server=True)
|