htrflow_mcp / app.py
Gabriel's picture
Update app.py
0914acb verified
import gradio as gr
import json
import tempfile
import os
import zipfile
import shutil
from typing import List, Optional, Literal, Union, Dict
from PIL import Image
import requests
from pathlib import Path
import spaces
from visualizer import htrflow_visualizer
from htrflow.volume.volume import Collection
from htrflow.pipeline.pipeline import Pipeline
DEFAULT_OUTPUT = "alto"
FORMAT_CHOICES = [
"letter_english",
"letter_swedish",
"spread_english",
"spread_swedish",
]
FILE_CHOICES = ["txt", "alto", "page", "json"]
FormatChoices = Literal[
"letter_english", "letter_swedish", "spread_english", "spread_swedish"
]
FileChoices = Literal["txt", "alto", "page", "json"]
PIPELINE_CONFIGS = {
"letter_english": {
"steps": [
{
"step": "Segmentation",
"settings": {
"model": "yolo",
"model_settings": {
"model": "Riksarkivet/yolov9-lines-within-regions-1"
},
"generation_settings": {"batch_size": 8},
},
},
{
"step": "TextRecognition",
"settings": {
"model": "TrOCR",
"model_settings": {"model": "microsoft/trocr-base-handwritten"},
"generation_settings": {"batch_size": 16},
},
},
{"step": "OrderLines"},
]
},
"letter_swedish": {
"steps": [
{
"step": "Segmentation",
"settings": {
"model": "yolo",
"model_settings": {
"model": "Riksarkivet/yolov9-lines-within-regions-1"
},
"generation_settings": {"batch_size": 8},
},
},
{
"step": "TextRecognition",
"settings": {
"model": "TrOCR",
"model_settings": {
"model": "Riksarkivet/trocr-base-handwritten-hist-swe-2"
},
"generation_settings": {"batch_size": 16},
},
},
{"step": "OrderLines"},
]
},
"spread_english": {
"steps": [
{
"step": "Segmentation",
"settings": {
"model": "yolo",
"model_settings": {"model": "Riksarkivet/yolov9-regions-1"},
"generation_settings": {"batch_size": 4},
},
},
{
"step": "Segmentation",
"settings": {
"model": "yolo",
"model_settings": {
"model": "Riksarkivet/yolov9-lines-within-regions-1"
},
"generation_settings": {"batch_size": 8},
},
},
{
"step": "TextRecognition",
"settings": {
"model": "TrOCR",
"model_settings": {"model": "microsoft/trocr-base-handwritten"},
"generation_settings": {"batch_size": 16},
},
},
{"step": "ReadingOrderMarginalia", "settings": {"two_page": True}},
]
},
"spread_swedish": {
"steps": [
{
"step": "Segmentation",
"settings": {
"model": "yolo",
"model_settings": {"model": "Riksarkivet/yolov9-regions-1"},
"generation_settings": {"batch_size": 4},
},
},
{
"step": "Segmentation",
"settings": {
"model": "yolo",
"model_settings": {
"model": "Riksarkivet/yolov9-lines-within-regions-1"
},
"generation_settings": {"batch_size": 8},
},
},
{
"step": "TextRecognition",
"settings": {
"model": "TrOCR",
"model_settings": {
"model": "Riksarkivet/trocr-base-handwritten-hist-swe-2"
},
"generation_settings": {"batch_size": 16},
},
},
{"step": "ReadingOrderMarginalia", "settings": {"two_page": True}},
]
},
}
def handle_image_input(image_path: Union[str, None], progress: gr.Progress = None, desc_prefix: str = "") -> str:
"""
Handle image input from various sources (local file, URL, or uploaded file).
Args:
image_path: Path to image file or URL
progress: Progress tracker for UI updates
desc_prefix: Prefix for progress descriptions
Returns:
Local file path to the image
"""
if not image_path:
raise ValueError("No image provided. Please upload an image or provide a URL.")
if progress:
progress(0.1, desc=f"{desc_prefix}Processing image input...")
# If it's a URL, download the image
if isinstance(image_path, str) and (image_path.startswith("http://") or image_path.startswith("https://")):
try:
if progress:
progress(0.2, desc=f"{desc_prefix}Downloading image from URL...")
response = requests.get(image_path, timeout=30)
response.raise_for_status()
# Save to temporary file
with tempfile.NamedTemporaryFile(delete=False, suffix=".jpg") as tmp_file:
tmp_file.write(response.content)
image_path = tmp_file.name
# Verify it's a valid image
try:
img = Image.open(image_path)
img.verify()
except Exception as e:
os.unlink(image_path)
raise ValueError(f"Downloaded file is not a valid image: {str(e)}")
except requests.RequestException as e:
raise ValueError(f"Failed to download image from URL: {str(e)}")
# Verify the file exists
if not os.path.exists(image_path):
raise ValueError(f"Image file not found: {image_path}")
return image_path
def parse_image_input(image_input: Union[str, List[str], None]) -> List[str]:
"""
Parse image input which can be a single path, multiple paths, or URLs separated by newlines.
Args:
image_input: Single image path, list of paths, or newline-separated URLs/paths
Returns:
List of image paths/URLs
"""
if not image_input:
return []
if isinstance(image_input, list):
return image_input
if isinstance(image_input, str):
# Check if it's multiple URLs/paths separated by newlines
lines = image_input.strip().split('\n')
paths = []
for line in lines:
line = line.strip()
if line: # Skip empty lines
paths.append(line)
return paths if paths else [image_input]
return []
@spaces.GPU
def _process_htr_pipeline_batch(
image_paths: List[str],
document_type: FormatChoices,
custom_settings: Optional[str] = None,
progress: gr.Progress = None
) -> Dict[str, Collection]:
"""Process HTR pipeline for multiple images and return processed collections."""
results = {}
temp_files = []
total_images = len(image_paths)
if custom_settings:
try:
config = json.loads(custom_settings)
except json.JSONDecodeError:
raise ValueError("Invalid JSON in custom_settings parameter. Please check your JSON syntax.")
else:
config = PIPELINE_CONFIGS[document_type]
# Initialize pipeline once for all images
pipeline = Pipeline.from_config(config)
for idx, image_path in enumerate(image_paths):
try:
image_name = Path(image_path).stem if not image_path.startswith("http") else f"image_{idx+1}"
if progress:
progress((idx + 0.2) / total_images,
desc=f"Processing image {idx+1}/{total_images}: {image_name}")
# Handle image input
processed_path = handle_image_input(image_path, progress,
desc_prefix=f"[{idx+1}/{total_images}] ")
# Track temp files for cleanup
if processed_path.startswith(tempfile.gettempdir()):
temp_files.append(processed_path)
if progress:
progress((idx + 0.5) / total_images,
desc=f"Running HTR on image {idx+1}/{total_images}: {image_name}")
# Process with pipeline
collection = Collection([processed_path])
processed_collection = pipeline.run(collection)
results[image_name] = processed_collection
if progress:
progress((idx + 1.0) / total_images,
desc=f"Completed image {idx+1}/{total_images}: {image_name}")
except Exception as e:
results[image_name] = f"Error: {str(e)}"
print(f"Error processing {image_path}: {str(e)}")
# Cleanup temp files
for temp_file in temp_files:
try:
os.unlink(temp_file)
except:
pass
if progress:
progress(1.0, desc=f"Completed processing all {total_images} images!")
return results
def extract_text_from_collection(collection: Collection) -> str:
"""Extract and combine text from all nodes in the collection."""
text_lines = []
for page in collection.pages:
for node in page.traverse():
if hasattr(node, "text") and node.text:
text_lines.append(node.text)
return "\n".join(text_lines)
def htr_text(
image_input: Union[str, List[str]],
document_type: FormatChoices = "letter_swedish",
custom_settings: Optional[str] = None,
return_format: str = "separate", # "separate" or "combined"
progress: gr.Progress = gr.Progress()
) -> str:
"""
Extract text from handwritten documents using HTR.
Handles both single images and multiple images.
Args:
image_input: Single image path/URL, multiple paths/URLs (newline-separated), or list of uploaded files
document_type: Type of document layout - choose based on your documents' structure and language
custom_settings: Optional JSON configuration for advanced pipeline customization
return_format: "separate" to show each document's text separately, "combined" to merge all text
progress: Progress tracker for UI updates
Returns:
Extracted text from all handwritten documents
"""
try:
if progress:
progress(0, desc="Starting HTR text extraction...")
# Parse input to get list of images
image_paths = parse_image_input(image_input)
if not image_paths:
return "No images provided. Please upload images or provide URLs."
# Adjust description based on single vs multiple
num_images = len(image_paths)
desc = f"Processing {num_images} image{'s' if num_images > 1 else ''}..."
if progress:
progress(0.1, desc=desc)
# Process all images
results = _process_htr_pipeline_batch(
image_paths, document_type, custom_settings, progress
)
# Extract text from results
all_texts = []
for image_name, collection in results.items():
if isinstance(collection, str): # Error case
all_texts.append(f"=== {image_name} ===\n{collection}\n")
else:
text = extract_text_from_collection(collection)
if return_format == "separate":
all_texts.append(f"=== {image_name} ===\n{text}\n")
else:
all_texts.append(text)
# Return formatted result
if return_format == "separate":
return "\n".join(all_texts)
else:
return "\n\n".join(all_texts)
except ValueError as e:
return f"Input error: {str(e)}"
except Exception as e:
return f"HTR text extraction failed: {str(e)}"
def htr_generate_files(
image_input: Union[str, List[str]],
document_type: FormatChoices = "letter_swedish",
output_format: FileChoices = DEFAULT_OUTPUT,
custom_settings: Optional[str] = None,
progress: gr.Progress = gr.Progress()
) -> str:
"""
Process handwritten documents and generate formatted output files.
Returns a ZIP file for multiple documents, or single file for single document.
Args:
image_input: Single image path/URL, multiple paths/URLs (newline-separated), or list of uploaded files
document_type: Type of document layout - affects segmentation and reading order
output_format: Desired output format (txt for plain text, alto/page for XML with coordinates, json for structured data)
custom_settings: Optional JSON configuration for advanced pipeline customization
progress: Progress tracker for UI updates
Returns:
Path to generated file(s)
"""
try:
if progress:
progress(0, desc="Starting HTR file processing...")
# Parse input to get list of images
image_paths = parse_image_input(image_input)
if not image_paths:
error_file = tempfile.NamedTemporaryFile(mode='w', delete=False, suffix='.txt')
error_file.write("No images provided. Please upload images or provide URLs.")
error_file.close()
return error_file.name
num_images = len(image_paths)
if progress:
progress(0.1, desc=f"Processing {num_images} image{'s' if num_images > 1 else ''}...")
# Process all images
results = _process_htr_pipeline_batch(
image_paths, document_type, custom_settings, progress
)
if progress:
progress(0.9, desc="Creating output files...")
# Create temporary directory for output files
temp_dir = Path(tempfile.mkdtemp())
output_files = []
for image_name, collection in results.items():
if isinstance(collection, str): # Error case
# Write error to text file
error_file_path = temp_dir / f"{image_name}_error.txt"
with open(error_file_path, 'w') as f:
f.write(collection)
output_files.append(error_file_path)
else:
# Save collection in requested format
export_dir = temp_dir / image_name
collection.save(directory=str(export_dir), serializer=output_format)
# Find and rename the generated file
for root, _, files in os.walk(export_dir):
for file in files:
old_path = Path(root) / file
file_ext = Path(file).suffix
new_filename = (
f"{image_name}.{output_format}"
if not file_ext
else f"{image_name}{file_ext}"
)
new_path = temp_dir / new_filename
shutil.move(str(old_path), str(new_path))
output_files.append(new_path)
break
# Return single file or ZIP based on input count
if len(output_files) == 1 and len(image_paths) == 1:
# Single file - return directly
if progress:
progress(1.0, desc="Processing complete!")
return str(output_files[0])
else:
# Multiple files - create ZIP
zip_path = temp_dir / f"htr_output_{output_format}.zip"
with zipfile.ZipFile(zip_path, 'w', zipfile.ZIP_DEFLATED) as zipf:
for file_path in output_files:
zipf.write(file_path, file_path.name)
if progress:
progress(1.0, desc=f"Processing complete! Generated {len(output_files)} files.")
return str(zip_path)
except ValueError as e:
error_file = tempfile.NamedTemporaryFile(mode='w', delete=False, suffix='.txt')
error_file.write(f"Input error: {str(e)}")
error_file.close()
return error_file.name
except Exception as e:
error_file = tempfile.NamedTemporaryFile(mode='w', delete=False, suffix='.txt')
error_file.write(f"HTR file generation failed: {str(e)}")
error_file.close()
return error_file.name
def htr_visualize(
image_input: Union[str, List[str]],
htr_documents: Union[List[str], None],
progress: gr.Progress = gr.Progress()
) -> str:
"""
Create visualizations for HTR results overlaid on original documents.
Returns a ZIP file for multiple documents, or single image for single document.
Args:
image_input: Original document image paths/URLs (newline-separated if string)
htr_documents: HTR output files (ALTO/PAGE XML) - must match order of images
progress: Progress tracker for UI updates
Returns:
Path to visualization file(s)
"""
try:
if progress:
progress(0, desc="Starting visualization generation...")
# Parse inputs
image_paths = parse_image_input(image_input)
# Handle htr_documents - it should be a list of file paths
if not htr_documents:
raise ValueError("No HTR documents provided")
# If htr_documents is a list of file objects from Gradio, extract paths
htr_paths = []
if isinstance(htr_documents, list):
for doc in htr_documents:
if isinstance(doc, str):
htr_paths.append(doc)
elif hasattr(doc, 'name'):
htr_paths.append(doc.name)
else:
htr_paths.append(str(doc))
else:
# Single file case
if isinstance(htr_documents, str):
htr_paths = [htr_documents]
elif hasattr(htr_documents, 'name'):
htr_paths = [htr_documents.name]
else:
htr_paths = [str(htr_documents)]
if not image_paths:
raise ValueError("No images provided")
if len(image_paths) != len(htr_paths):
raise ValueError(f"Number of images ({len(image_paths)}) doesn't match number of HTR documents ({len(htr_paths)})")
num_docs = len(image_paths)
if progress:
progress(0.1, desc=f"Creating visualization{'s' if num_docs > 1 else ''} for {num_docs} document{'s' if num_docs > 1 else ''}...")
temp_dir = Path(tempfile.mkdtemp())
output_files = []
temp_files = []
for idx, (image_path, htr_path) in enumerate(zip(image_paths, htr_paths)):
try:
image_name = Path(image_path).stem if not image_path.startswith("http") else f"image_{idx+1}"
if progress:
progress((idx + 0.3) / num_docs,
desc=f"Visualizing document {idx+1}/{num_docs}: {image_name}")
# Handle image input
processed_image = handle_image_input(image_path, progress,
desc_prefix=f"[{idx+1}/{num_docs}] ")
if processed_image.startswith(tempfile.gettempdir()):
temp_files.append(processed_image)
# Generate visualization - use the last parameter for output path
output_viz_path = str(temp_dir / f"{image_name}_visualization.png")
viz_result = htrflow_visualizer(processed_image, htr_path, output_viz_path)
# Check if visualization was created
if os.path.exists(output_viz_path):
output_files.append(Path(output_viz_path))
elif viz_result and os.path.exists(viz_result):
# Fallback: if viz_result points to a different file
viz_path = temp_dir / f"{image_name}_visualization.png"
shutil.move(viz_result, str(viz_path))
output_files.append(viz_path)
else:
raise ValueError("Visualization generation failed - no output file created")
except Exception as e:
# Create error file for this visualization
error_path = temp_dir / f"{image_name}_viz_error.txt"
with open(error_path, 'w') as f:
f.write(f"Visualization failed: {str(e)}")
output_files.append(error_path)
print(f"Error visualizing {image_name}: {str(e)}")
# Cleanup temp files
for temp_file in temp_files:
try:
os.unlink(temp_file)
except:
pass
# Return single file or ZIP based on input count
if len(output_files) == 1 and num_docs == 1:
# Single visualization - return directly
if progress:
progress(1.0, desc="Visualization complete!")
return str(output_files[0])
else:
# Multiple visualizations - create ZIP
if progress:
progress(0.9, desc="Creating ZIP archive...")
zip_path = temp_dir / "htr_visualizations.zip"
with zipfile.ZipFile(zip_path, 'w', zipfile.ZIP_DEFLATED) as zipf:
for file_path in output_files:
zipf.write(file_path, file_path.name)
if progress:
progress(1.0, desc=f"Visualization complete! Created {len(output_files)} visualization{'s' if len(output_files) > 1 else ''}.")
return str(zip_path)
except Exception as e:
error_file = tempfile.NamedTemporaryFile(mode='w', delete=False, suffix='.txt')
error_file.write(f"Visualization failed: {str(e)}")
error_file.close()
return error_file.name
def create_htrflow_mcp_server():
# HTR Text extraction interface
htr_text_interface = gr.Interface(
fn=htr_text,
inputs=[
gr.Textbox(
label="Image Input",
placeholder="Single image path/URL or multiple (one per line)\nYou can also drag and drop files here",
lines=3
),
gr.Dropdown(
choices=FORMAT_CHOICES,
value="letter_swedish",
label="Document Type",
info="Select the type that best matches your documents' layout and language"
),
gr.Textbox(
label="Custom Settings (JSON)",
placeholder='{"steps": [...]} - Leave empty for default settings',
value="",
lines=3
),
gr.Radio(
choices=["separate", "combined"],
value="separate",
label="Output Format",
info="'separate' shows each document's text with headers, 'combined' merges all text"
),
],
outputs=[gr.Textbox(label="Extracted Text", lines=20)],
title="Extract Text from Handwritten Documents",
description="Process one or more handwritten document images. Works with letters and book spreads in English and Swedish.",
api_name="htr_text",
)
# HTR File generation interface
htr_files_interface = gr.Interface(
fn=htr_generate_files,
inputs=[
gr.Textbox(
label="Image Input",
placeholder="Single image path/URL or multiple (one per line)\nYou can also drag and drop files here",
lines=3
),
gr.Dropdown(
choices=FORMAT_CHOICES,
value="letter_swedish",
label="Document Type",
info="Select the type that best matches your documents' layout and language"
),
gr.Dropdown(
choices=FILE_CHOICES,
value=DEFAULT_OUTPUT,
label="Output Format",
info="ALTO/PAGE: XML with coordinates | JSON: Structured data | TXT: Plain text only"
),
gr.Textbox(
label="Custom Settings (JSON)",
placeholder='{"steps": [...]} - Leave empty for default settings',
value="",
lines=3
),
],
outputs=[gr.File(label="Download HTR Output")],
title="Generate HTR Output Files",
description="Process handwritten documents and export in various formats. Returns ZIP for multiple files.",
api_name="htr_generate_files",
)
# HTR Visualization interface
htr_viz_interface = gr.Interface(
fn=htr_visualize,
inputs=[
gr.Textbox(
label="Original Image Paths/URLs",
placeholder="One path/URL per line",
lines=3
),
gr.File(
label="Upload HTR XML Files (ALTO/PAGE)",
file_types=[".xml"],
file_count="multiple"
),
],
outputs=gr.File(label="Download Visualization"),
title="Visualize HTR Results",
description="Create annotated images showing detected regions and text. Files must be in matching order.",
api_name="htr_visualize",
)
# Create tabbed interface
demo = gr.TabbedInterface(
[
htr_text_interface,
htr_files_interface,
htr_viz_interface,
],
[
"📚 Extract Text",
"📁 Generate Files",
"🖼️ Visualize Results",
],
title="🖋️ HTRflow - Handwritten Text Recognition",
analytics_enabled=False,
)
return demo
if __name__ == "__main__":
demo = create_htrflow_mcp_server()
demo.launch(
mcp_server=True,
share=False,
debug=False,
show_api=True,
favicon_path=None,
)