Spaces:
Running
Running
Upload 2 files
Browse files- app.py +574 -0
- requirements.txt +8 -0
app.py
ADDED
@@ -0,0 +1,574 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
import os
|
3 |
+
import time
|
4 |
+
import torch
|
5 |
+
import tempfile
|
6 |
+
from PIL import Image
|
7 |
+
from dotenv import load_dotenv
|
8 |
+
import logging
|
9 |
+
from datetime import datetime
|
10 |
+
|
11 |
+
# Set up logging
|
12 |
+
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
|
13 |
+
logger = logging.getLogger(__name__)
|
14 |
+
|
15 |
+
# Load environment variables
|
16 |
+
load_dotenv()
|
17 |
+
HF_TOKEN = os.getenv("HF_TOKEN")
|
18 |
+
CACHE_DIR = os.getenv("CACHE_DIR", os.path.join(tempfile.gettempdir(), "smoldocling_cache"))
|
19 |
+
|
20 |
+
# Ensure cache directory exists
|
21 |
+
os.makedirs(CACHE_DIR, exist_ok=True)
|
22 |
+
|
23 |
+
# Import for Transformers approach
|
24 |
+
try:
|
25 |
+
from transformers import AutoProcessor, AutoModelForVision2Seq
|
26 |
+
from huggingface_hub import login
|
27 |
+
transformers_available = True
|
28 |
+
except ImportError:
|
29 |
+
transformers_available = False
|
30 |
+
|
31 |
+
try:
|
32 |
+
from docling_core.types.doc import DoclingDocument
|
33 |
+
from docling_core.types.doc.document import DocTagsDocument
|
34 |
+
docling_available = True
|
35 |
+
except ImportError:
|
36 |
+
docling_available = False
|
37 |
+
|
38 |
+
# Global variables for model caching
|
39 |
+
processor = None
|
40 |
+
model = None
|
41 |
+
|
42 |
+
def check_dependencies():
|
43 |
+
"""Check if all required dependencies are installed"""
|
44 |
+
missing = []
|
45 |
+
if not transformers_available:
|
46 |
+
missing.append("transformers huggingface_hub")
|
47 |
+
if not docling_available:
|
48 |
+
missing.append("docling-core")
|
49 |
+
|
50 |
+
return missing
|
51 |
+
|
52 |
+
def get_available_devices():
|
53 |
+
"""Get available processing devices"""
|
54 |
+
devices = ["cpu"]
|
55 |
+
if torch.cuda.is_available():
|
56 |
+
cuda_count = torch.cuda.device_count()
|
57 |
+
for i in range(cuda_count):
|
58 |
+
devices.append(f"cuda:{i} ({torch.cuda.get_device_name(i)})")
|
59 |
+
return devices
|
60 |
+
|
61 |
+
def get_device_from_selection(selection):
|
62 |
+
"""Convert user-friendly device selection to torch device"""
|
63 |
+
if selection.startswith("cuda:"):
|
64 |
+
return selection.split(" ")[0] # Extract just the "cuda:X" part
|
65 |
+
return "cpu"
|
66 |
+
|
67 |
+
@st.cache_resource
|
68 |
+
def load_model(_device):
|
69 |
+
"""Load and cache the model to avoid reloading"""
|
70 |
+
global processor, model
|
71 |
+
|
72 |
+
# Authenticate with Hugging Face
|
73 |
+
if HF_TOKEN:
|
74 |
+
login(token=HF_TOKEN)
|
75 |
+
|
76 |
+
try:
|
77 |
+
logger.info(f"Loading SmolDocling model on {_device}...")
|
78 |
+
processor = AutoProcessor.from_pretrained(
|
79 |
+
"ds4sd/SmolDocling-256M-preview",
|
80 |
+
cache_dir=CACHE_DIR
|
81 |
+
)
|
82 |
+
model = AutoModelForVision2Seq.from_pretrained(
|
83 |
+
"ds4sd/SmolDocling-256M-preview",
|
84 |
+
torch_dtype=torch.float16 if _device.startswith("cuda") else torch.float32,
|
85 |
+
cache_dir=CACHE_DIR
|
86 |
+
).to(_device)
|
87 |
+
logger.info("Model loaded successfully")
|
88 |
+
return processor, model
|
89 |
+
except Exception as e:
|
90 |
+
logger.error(f"Error loading model: {str(e)}")
|
91 |
+
raise
|
92 |
+
|
93 |
+
def optimize_image(image, max_size=1600):
|
94 |
+
"""Optimize image size while maintaining aspect ratio"""
|
95 |
+
width, height = image.size
|
96 |
+
if max(width, height) > max_size:
|
97 |
+
if width > height:
|
98 |
+
new_width = max_size
|
99 |
+
new_height = int(height * (max_size / width))
|
100 |
+
else:
|
101 |
+
new_height = max_size
|
102 |
+
new_width = int(width * (max_size / height))
|
103 |
+
image = image.resize((new_width, new_height), Image.LANCZOS)
|
104 |
+
return image
|
105 |
+
|
106 |
+
def process_single_image(image, prompt_text="Convert this page to docling.", device="cpu", show_progress=None):
|
107 |
+
"""Process a single image"""
|
108 |
+
global processor, model
|
109 |
+
|
110 |
+
# Optimize image
|
111 |
+
image = optimize_image(image)
|
112 |
+
|
113 |
+
start_time = time.time()
|
114 |
+
|
115 |
+
# Load the model if not already loaded
|
116 |
+
processor, model = load_model(device)
|
117 |
+
|
118 |
+
# Create input messages
|
119 |
+
messages = [
|
120 |
+
{
|
121 |
+
"role": "user",
|
122 |
+
"content": [
|
123 |
+
{"type": "image"},
|
124 |
+
{"type": "text", "text": prompt_text}
|
125 |
+
]
|
126 |
+
},
|
127 |
+
]
|
128 |
+
|
129 |
+
# Prepare inputs
|
130 |
+
prompt = processor.apply_chat_template(messages, add_generation_prompt=True)
|
131 |
+
inputs = processor(text=prompt, images=[image], return_tensors="pt")
|
132 |
+
inputs = inputs.to(device)
|
133 |
+
|
134 |
+
# Generate outputs
|
135 |
+
with torch.no_grad(): # Add this to save memory
|
136 |
+
generated_ids = model.generate(
|
137 |
+
**inputs,
|
138 |
+
max_new_tokens=1500, # Increased for better results
|
139 |
+
do_sample=False, # Deterministic generation
|
140 |
+
num_beams=1, # Simple beam search
|
141 |
+
temperature=1.0, # No temperature scaling
|
142 |
+
)
|
143 |
+
|
144 |
+
prompt_length = inputs.input_ids.shape[1]
|
145 |
+
trimmed_generated_ids = generated_ids[:, prompt_length:]
|
146 |
+
doctags = processor.batch_decode(
|
147 |
+
trimmed_generated_ids,
|
148 |
+
skip_special_tokens=False,
|
149 |
+
)[0].lstrip()
|
150 |
+
|
151 |
+
# Clean the output
|
152 |
+
doctags = doctags.replace("<end_of_utterance>", "").strip()
|
153 |
+
|
154 |
+
# Populate document
|
155 |
+
doctags_doc = DocTagsDocument.from_doctags_and_image_pairs([doctags], [image])
|
156 |
+
|
157 |
+
# Create a docling document
|
158 |
+
doc = DoclingDocument(name="Document")
|
159 |
+
doc.load_from_doctags(doctags_doc)
|
160 |
+
|
161 |
+
# Export as markdown
|
162 |
+
md_content = doc.export_to_markdown()
|
163 |
+
|
164 |
+
# Export as HTML
|
165 |
+
html_content = doc.export_to_html()
|
166 |
+
|
167 |
+
# Get plain text
|
168 |
+
plain_text = doc.export_to_text()
|
169 |
+
|
170 |
+
processing_time = time.time() - start_time
|
171 |
+
|
172 |
+
return {
|
173 |
+
"doctags": doctags,
|
174 |
+
"markdown": md_content,
|
175 |
+
"html": html_content,
|
176 |
+
"text": plain_text,
|
177 |
+
"processing_time": processing_time
|
178 |
+
}
|
179 |
+
|
180 |
+
def process_batch(images, prompt_text, device, progress_bar=None):
|
181 |
+
"""Process a batch of images with progress tracking"""
|
182 |
+
results = []
|
183 |
+
total = len(images)
|
184 |
+
|
185 |
+
for idx, image in enumerate(images):
|
186 |
+
if progress_bar:
|
187 |
+
progress_bar.progress((idx) / total, text=f"Processing image {idx+1}/{total}")
|
188 |
+
|
189 |
+
result = process_single_image(image, prompt_text, device)
|
190 |
+
results.append(result)
|
191 |
+
|
192 |
+
if progress_bar:
|
193 |
+
progress_bar.progress((idx + 1) / total, text=f"Processed {idx+1}/{total} images")
|
194 |
+
|
195 |
+
return results
|
196 |
+
|
197 |
+
def save_session_history(results):
|
198 |
+
"""Save processing results to session history"""
|
199 |
+
if 'history' not in st.session_state:
|
200 |
+
st.session_state.history = []
|
201 |
+
|
202 |
+
timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
|
203 |
+
|
204 |
+
for idx, result in enumerate(results):
|
205 |
+
st.session_state.history.append({
|
206 |
+
"id": len(st.session_state.history) + 1,
|
207 |
+
"timestamp": timestamp,
|
208 |
+
"type": "Image " + str(idx + 1),
|
209 |
+
"processing_time": result["processing_time"],
|
210 |
+
"result": result
|
211 |
+
})
|
212 |
+
|
213 |
+
def display_history():
|
214 |
+
"""Display session history"""
|
215 |
+
if 'history' not in st.session_state or not st.session_state.history:
|
216 |
+
st.info("No processing history available")
|
217 |
+
return
|
218 |
+
|
219 |
+
st.subheader("Processing History")
|
220 |
+
|
221 |
+
for item in reversed(st.session_state.history):
|
222 |
+
with st.expander(f"#{item['id']} - {item['type']} ({item['timestamp']})"):
|
223 |
+
st.write(f"Processing time: {item['processing_time']:.2f} seconds")
|
224 |
+
tabs = st.tabs(["Markdown", "Text", "DocTags", "HTML"])
|
225 |
+
|
226 |
+
with tabs[0]:
|
227 |
+
st.markdown(item['result']['markdown'])
|
228 |
+
st.download_button(
|
229 |
+
"Download Markdown",
|
230 |
+
item['result']['markdown'],
|
231 |
+
file_name=f"output_{item['id']}.md"
|
232 |
+
)
|
233 |
+
|
234 |
+
with tabs[1]:
|
235 |
+
st.text_area("Plain Text", item['result']['text'], height=200)
|
236 |
+
st.download_button(
|
237 |
+
"Download Text",
|
238 |
+
item['result']['text'],
|
239 |
+
file_name=f"output_{item['id']}.txt"
|
240 |
+
)
|
241 |
+
|
242 |
+
with tabs[2]:
|
243 |
+
st.text_area("DocTags", item['result']['doctags'], height=200)
|
244 |
+
st.download_button(
|
245 |
+
"Download DocTags",
|
246 |
+
item['result']['doctags'],
|
247 |
+
file_name=f"output_{item['id']}.dt"
|
248 |
+
)
|
249 |
+
|
250 |
+
with tabs[3]:
|
251 |
+
st.code(item['result']['html'], language="html")
|
252 |
+
st.download_button(
|
253 |
+
"Download HTML",
|
254 |
+
item['result']['html'],
|
255 |
+
file_name=f"output_{item['id']}.html"
|
256 |
+
)
|
257 |
+
|
258 |
+
def main():
|
259 |
+
# App configuration
|
260 |
+
st.set_page_config(
|
261 |
+
page_title="SmolDocling OCR App",
|
262 |
+
page_icon="📄",
|
263 |
+
layout="wide",
|
264 |
+
initial_sidebar_state="expanded"
|
265 |
+
)
|
266 |
+
|
267 |
+
# Custom theme
|
268 |
+
st.markdown("""
|
269 |
+
<style>
|
270 |
+
.main-header {
|
271 |
+
font-size: 2.5rem;
|
272 |
+
margin-bottom: 0.5rem;
|
273 |
+
}
|
274 |
+
.sub-header {
|
275 |
+
font-size: 1.2rem;
|
276 |
+
color: #666;
|
277 |
+
margin-bottom: 2rem;
|
278 |
+
}
|
279 |
+
.stTabs [data-baseweb="tab-list"] {
|
280 |
+
gap: 2px;
|
281 |
+
}
|
282 |
+
.stTabs [data-baseweb="tab"] {
|
283 |
+
padding: 10px 16px;
|
284 |
+
background-color: #f0f2f6;
|
285 |
+
}
|
286 |
+
.stTabs [aria-selected="true"] {
|
287 |
+
background-color: #e6f0ff;
|
288 |
+
}
|
289 |
+
</style>
|
290 |
+
""", unsafe_allow_html=True)
|
291 |
+
|
292 |
+
# App header
|
293 |
+
st.markdown('<p class="main-header">SmolDocling OCR App</p>', unsafe_allow_html=True)
|
294 |
+
st.markdown('<p class="sub-header">Extract text from images using SmolDocling AI</p>', unsafe_allow_html=True)
|
295 |
+
|
296 |
+
# Check dependencies
|
297 |
+
missing_deps = check_dependencies()
|
298 |
+
if missing_deps:
|
299 |
+
st.error(f"Missing dependencies: {', '.join(missing_deps)}. Please install them to use this app.")
|
300 |
+
st.info("Install with: pip install " + " ".join(missing_deps))
|
301 |
+
st.stop()
|
302 |
+
|
303 |
+
# Initialize session state
|
304 |
+
if 'results' not in st.session_state:
|
305 |
+
st.session_state.results = []
|
306 |
+
|
307 |
+
# Create sidebar
|
308 |
+
with st.sidebar:
|
309 |
+
st.header("Configuration")
|
310 |
+
|
311 |
+
# Device selection
|
312 |
+
st.subheader("Processing Device")
|
313 |
+
available_devices = get_available_devices()
|
314 |
+
selected_device = st.selectbox(
|
315 |
+
"Select processing device",
|
316 |
+
available_devices,
|
317 |
+
index=0 if len(available_devices) == 1 else 1, # Default to CUDA if available
|
318 |
+
help="Choose the device for model inference. GPU (CUDA) is recommended for faster processing."
|
319 |
+
)
|
320 |
+
device = get_device_from_selection(selected_device)
|
321 |
+
|
322 |
+
# Model info
|
323 |
+
st.info(f"Selected device: {selected_device}")
|
324 |
+
|
325 |
+
if device == "cpu":
|
326 |
+
st.warning("⚠️ CPU processing may be slow. Select a GPU device if available for faster performance.")
|
327 |
+
|
328 |
+
# Memory management
|
329 |
+
if device.startswith("cuda"):
|
330 |
+
with st.expander("GPU Memory Management"):
|
331 |
+
st.write("Current GPU Memory Usage:")
|
332 |
+
if torch.cuda.is_available():
|
333 |
+
gpu_idx = int(device.split(":")[1]) if ":" in device else 0
|
334 |
+
allocated = torch.cuda.memory_allocated(gpu_idx) / (1024 ** 3)
|
335 |
+
reserved = torch.cuda.memory_reserved(gpu_idx) / (1024 ** 3)
|
336 |
+
st.progress(allocated / (torch.cuda.get_device_properties(gpu_idx).total_memory / (1024 ** 3)))
|
337 |
+
st.write(f"Allocated: {allocated:.2f} GB")
|
338 |
+
st.write(f"Reserved: {reserved:.2f} GB")
|
339 |
+
|
340 |
+
if st.button("Clear GPU Cache"):
|
341 |
+
torch.cuda.empty_cache()
|
342 |
+
st.success("GPU cache cleared")
|
343 |
+
|
344 |
+
# Upload options
|
345 |
+
st.subheader("Upload Options")
|
346 |
+
upload_option = st.radio("Choose upload option:", ["Single Image", "Multiple Images"])
|
347 |
+
|
348 |
+
# Advanced options
|
349 |
+
with st.expander("Advanced Options"):
|
350 |
+
task_type = st.selectbox(
|
351 |
+
"Select task type",
|
352 |
+
[
|
353 |
+
"Convert this page to docling.",
|
354 |
+
"Convert this table to OTSL.",
|
355 |
+
"Convert code to text.",
|
356 |
+
"Convert formula to latex.",
|
357 |
+
"Convert chart to OTSL.",
|
358 |
+
"Extract all section header elements on the page."
|
359 |
+
]
|
360 |
+
)
|
361 |
+
|
362 |
+
custom_prompt = st.text_area(
|
363 |
+
"Custom prompt (optional)",
|
364 |
+
value="",
|
365 |
+
help="Provide a custom prompt if needed. Leave empty to use the selected task type."
|
366 |
+
)
|
367 |
+
|
368 |
+
max_image_size = st.slider(
|
369 |
+
"Max image dimension (pixels)",
|
370 |
+
min_value=800,
|
371 |
+
max_value=3200,
|
372 |
+
value=1600,
|
373 |
+
step=100,
|
374 |
+
help="Larger values may improve OCR quality but use more memory"
|
375 |
+
)
|
376 |
+
|
377 |
+
final_prompt = custom_prompt if custom_prompt else task_type
|
378 |
+
|
379 |
+
# Upload controls
|
380 |
+
st.subheader("Upload Image(s)")
|
381 |
+
if upload_option == "Single Image":
|
382 |
+
uploaded_file = st.file_uploader("Upload image", type=["jpg", "jpeg", "png", "pdf"])
|
383 |
+
|
384 |
+
if uploaded_file is not None:
|
385 |
+
try:
|
386 |
+
image = Image.open(uploaded_file).convert("RGB")
|
387 |
+
st.image(image, caption="Uploaded Image", width=250)
|
388 |
+
except Exception as e:
|
389 |
+
st.error(f"Error loading image: {str(e)}")
|
390 |
+
else:
|
391 |
+
uploaded_files = st.file_uploader(
|
392 |
+
"Upload multiple images",
|
393 |
+
type=["jpg", "jpeg", "png"],
|
394 |
+
accept_multiple_files=True
|
395 |
+
)
|
396 |
+
|
397 |
+
if uploaded_files:
|
398 |
+
st.success(f"{len(uploaded_files)} images uploaded")
|
399 |
+
|
400 |
+
# Process button
|
401 |
+
if (upload_option == "Single Image" and 'uploaded_file' in locals() and uploaded_file is not None) or \
|
402 |
+
(upload_option == "Multiple Images" and 'uploaded_files' in locals() and uploaded_files):
|
403 |
+
process_button = st.button("Process Image(s)", type="primary")
|
404 |
+
|
405 |
+
# History button
|
406 |
+
st.subheader("History")
|
407 |
+
if st.button("Show Processing History"):
|
408 |
+
st.session_state.show_history = True
|
409 |
+
|
410 |
+
# About section
|
411 |
+
with st.expander("About SmolDocling OCR"):
|
412 |
+
st.write("""
|
413 |
+
This app uses SmolDocling, a powerful OCR model for document understanding from Hugging Face Hub.
|
414 |
+
|
415 |
+
The app extracts DocTags format and converts it to Markdown, HTML, and plain text for easy reading.
|
416 |
+
|
417 |
+
Available tasks:
|
418 |
+
- Convert pages to DocTags (general OCR)
|
419 |
+
- Convert tables to OTSL
|
420 |
+
- Convert code snippets to text
|
421 |
+
- Convert formulas to LaTeX
|
422 |
+
- Convert charts to OTSL
|
423 |
+
- Extract section headers
|
424 |
+
""")
|
425 |
+
|
426 |
+
# Main content area
|
427 |
+
if 'show_history' in st.session_state and st.session_state.show_history:
|
428 |
+
display_history()
|
429 |
+
st.session_state.show_history = False
|
430 |
+
elif upload_option == "Single Image" and 'uploaded_file' in locals() and uploaded_file is not None and process_button:
|
431 |
+
with st.spinner("Processing image..."):
|
432 |
+
try:
|
433 |
+
progress_bar = st.progress(0, text="Preparing to process...")
|
434 |
+
|
435 |
+
# Update global optimization settings
|
436 |
+
optimize_image.func_defaults = (max_image_size,)
|
437 |
+
|
438 |
+
result = process_single_image(image, final_prompt, device)
|
439 |
+
st.session_state.results = [result]
|
440 |
+
|
441 |
+
# Save to history
|
442 |
+
save_session_history(st.session_state.results)
|
443 |
+
|
444 |
+
progress_bar.progress(1.0, text="Processing complete!")
|
445 |
+
|
446 |
+
# Display results
|
447 |
+
tabs = st.tabs(["Markdown", "Text", "DocTags", "HTML"])
|
448 |
+
|
449 |
+
with tabs[0]:
|
450 |
+
st.subheader("Markdown Output")
|
451 |
+
st.markdown(result["markdown"])
|
452 |
+
st.download_button(
|
453 |
+
"Download Markdown",
|
454 |
+
result["markdown"],
|
455 |
+
file_name="output.md"
|
456 |
+
)
|
457 |
+
|
458 |
+
with tabs[1]:
|
459 |
+
st.subheader("Plain Text Output")
|
460 |
+
st.text_area("Extracted Text", result["text"], height=300)
|
461 |
+
st.download_button(
|
462 |
+
"Download Text",
|
463 |
+
result["text"],
|
464 |
+
file_name="output.txt"
|
465 |
+
)
|
466 |
+
|
467 |
+
with tabs[2]:
|
468 |
+
st.subheader("DocTags Output")
|
469 |
+
st.text_area("DocTags", result["doctags"], height=300)
|
470 |
+
st.download_button(
|
471 |
+
"Download DocTags",
|
472 |
+
result["doctags"],
|
473 |
+
file_name="output.dt"
|
474 |
+
)
|
475 |
+
|
476 |
+
with tabs[3]:
|
477 |
+
st.subheader("HTML Output")
|
478 |
+
st.code(result["html"], language="html")
|
479 |
+
st.download_button(
|
480 |
+
"Download HTML",
|
481 |
+
result["html"],
|
482 |
+
file_name="output.html"
|
483 |
+
)
|
484 |
+
|
485 |
+
st.success(f"Processing completed in {result['processing_time']:.2f} seconds on {selected_device}")
|
486 |
+
except Exception as e:
|
487 |
+
st.error(f"Error processing image: {str(e)}")
|
488 |
+
logger.error(f"Error processing image: {str(e)}", exc_info=True)
|
489 |
+
|
490 |
+
elif upload_option == "Multiple Images" and 'uploaded_files' in locals() and uploaded_files and process_button:
|
491 |
+
try:
|
492 |
+
images = [Image.open(file).convert("RGB") for file in uploaded_files]
|
493 |
+
|
494 |
+
if len(images) > 0:
|
495 |
+
with st.spinner(f"Processing {len(images)} images..."):
|
496 |
+
progress_bar = st.progress(0, text="Preparing to process...")
|
497 |
+
|
498 |
+
# Update global optimization settings
|
499 |
+
optimize_image.func_defaults = (max_image_size,)
|
500 |
+
|
501 |
+
results = process_batch(images, final_prompt, device, progress_bar)
|
502 |
+
st.session_state.results = results
|
503 |
+
|
504 |
+
# Save to history
|
505 |
+
save_session_history(results)
|
506 |
+
|
507 |
+
progress_bar.progress(1.0, text="Processing complete!")
|
508 |
+
|
509 |
+
# Display results
|
510 |
+
st.subheader("Processing Results")
|
511 |
+
|
512 |
+
total_time = sum(result["processing_time"] for result in results)
|
513 |
+
avg_time = total_time / len(results)
|
514 |
+
|
515 |
+
st.write(f"Total processing time: {total_time:.2f} seconds on {selected_device}")
|
516 |
+
st.write(f"Average processing time: {avg_time:.2f} seconds per image")
|
517 |
+
|
518 |
+
# Create tabs for each image
|
519 |
+
for idx, (result, image) in enumerate(zip(results, images)):
|
520 |
+
with st.expander(f"Image {idx+1} Results"):
|
521 |
+
col1, col2 = st.columns([1, 2])
|
522 |
+
|
523 |
+
with col1:
|
524 |
+
st.image(image, caption=f"Image {idx+1}", width=250)
|
525 |
+
st.write(f"Processing time: {result['processing_time']:.2f} seconds")
|
526 |
+
|
527 |
+
with col2:
|
528 |
+
inner_tabs = st.tabs(["Markdown", "Text", "DocTags", "HTML"])
|
529 |
+
|
530 |
+
with inner_tabs[0]:
|
531 |
+
st.markdown(result["markdown"])
|
532 |
+
st.download_button(
|
533 |
+
f"Download Markdown",
|
534 |
+
result["markdown"],
|
535 |
+
file_name=f"output_{idx+1}.md"
|
536 |
+
)
|
537 |
+
|
538 |
+
with inner_tabs[1]:
|
539 |
+
st.text_area("Plain Text", result["text"], height=200)
|
540 |
+
st.download_button(
|
541 |
+
f"Download Text",
|
542 |
+
result["text"],
|
543 |
+
file_name=f"output_{idx+1}.txt"
|
544 |
+
)
|
545 |
+
|
546 |
+
with inner_tabs[2]:
|
547 |
+
st.text_area("DocTags", result["doctags"], height=200)
|
548 |
+
st.download_button(
|
549 |
+
f"Download DocTags",
|
550 |
+
result["doctags"],
|
551 |
+
file_name=f"output_{idx+1}.dt"
|
552 |
+
)
|
553 |
+
|
554 |
+
with inner_tabs[3]:
|
555 |
+
st.code(result["html"], language="html")
|
556 |
+
st.download_button(
|
557 |
+
f"Download HTML",
|
558 |
+
result["html"],
|
559 |
+
file_name=f"output_{idx+1}.html"
|
560 |
+
)
|
561 |
+
|
562 |
+
st.success(f"All images processed successfully")
|
563 |
+
except Exception as e:
|
564 |
+
st.error(f"Error processing images: {str(e)}")
|
565 |
+
logger.error(f"Error processing images: {str(e)}", exc_info=True)
|
566 |
+
|
567 |
+
# Display a welcome message if no image has been uploaded
|
568 |
+
if ('uploaded_file' not in locals() or uploaded_file is None) and \
|
569 |
+
('uploaded_files' not in locals() or not uploaded_files):
|
570 |
+
st.info("👈 Upload an image using the sidebar to get started")
|
571 |
+
|
572 |
+
|
573 |
+
if __name__ == "__main__":
|
574 |
+
main()
|
requirements.txt
ADDED
@@ -0,0 +1,8 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
streamlit
|
2 |
+
torch
|
3 |
+
accelerate
|
4 |
+
transformers
|
5 |
+
docling-core
|
6 |
+
huggingface_hub
|
7 |
+
Pillow
|
8 |
+
python-dotenv
|