File size: 21,675 Bytes
1a7ee60 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 | """
Dataset Loading and Preprocessing for IDP Training
Supports CORD-v2, SROIE, and FUNSD datasets from Local Archives and Hugging Face
Converts to formats suitable for classification and NER training
"""
import io
import json
import logging
import os
import glob
from typing import Dict, List, Optional, Tuple, Union
import numpy as np
from datasets import load_dataset, Dataset
from PIL import Image
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
class CORDDatasetLoader:
"""Load and process CORD-v2 dataset from local archive"""
def __init__(self, base_path: str):
self.base_path = base_path
def load_dataset_splits(self):
"""Load CORD-v2 dataset from local files"""
logger.info(f"Loading CORD-v2 dataset from {self.base_path}...")
splits = {}
for split_name in ["train", "dev", "test"]:
# Map 'val' to 'dev' if needed, but directory is 'dev'
dir_name = "dev" if split_name == "validation" else split_name
if split_name == "val": dir_name = "dev"
json_dir = os.path.join(self.base_path, dir_name, "json")
if not os.path.exists(json_dir):
logger.warning(f"Split directory not found: {json_dir}")
continue
data = []
try:
files = os.listdir(json_dir)
json_files = [os.path.join(json_dir, f) for f in files if f.endswith('.json')]
except Exception as e:
logger.warning(f"Error listing dir {json_dir}: {e}")
json_files = []
for json_file in json_files:
try:
with open(json_file, 'r', encoding='utf-8') as f:
content = json.load(f)
# Add filename as id
content['id'] = os.path.basename(json_file).replace('.json', '')
data.append(content)
except Exception as e:
logger.warning(f"Error reading {json_file}: {e}")
splits[split_name] = data
logger.info(f"Loaded {len(data)} examples for split '{split_name}'")
return splits
@staticmethod
def extract_classification_data(dataset_split) -> List[Dict]:
"""
Extract data for document classification
"""
classification_data = []
for example in dataset_split:
try:
# CORD local JSON structure
text_lines = []
if "valid_line" in example:
for line in example["valid_line"]:
for word in line.get("words", []):
if "text" in word:
text_lines.append(word["text"])
text = " ".join(text_lines)
classification_data.append(
{
"text": text,
"label": "RECEIPT",
"image": None, # Image loading not implemented for local yet
"doc_id": example.get("id", ""),
}
)
except Exception as e:
doc_id = example.get("id", "unknown")
logger.warning(
f"Skipping record {doc_id} in CORD due to parsing error: {e}"
)
continue
return classification_data
@staticmethod
def extract_ner_data(dataset_split) -> List[Dict]:
"""
Extract data for NER training from local CORD JSON
"""
ner_data = []
# Mapping CORD fields to our entity types
entity_mapping = {
"menu.nm": "VENDOR_NAME", # Sometimes menu name is used as vendor/item
"menu.nm": "O", # Actually menu items are not usually vendor names in general receipt NER, but let's keep consistent with previous logic if possible.
# Previous logic: "menu.nm": "VENDOR_NAME". Wait, menu.nm is usually the item name.
# Let's map strictly important fields.
"total.total_price": "TOTAL_AMOUNT",
"total.tax_price": "TAX_AMOUNT",
"sub_total.subtotal_price": "TOTAL_AMOUNT",
# "menu.price": "TOTAL_AMOUNT", # Individual prices are not total
}
# Refined mapping based on CORD categories
# CORD categories: menu.nm, menu.cnt, menu.price, sub_total.subtotal_price, total.total_price, etc.
for example in dataset_split:
if "valid_line" not in example:
continue
tokens = []
labels = []
for line in example["valid_line"]:
category = line.get("category", "O")
# Map category to our label
# We need to be careful. CORD has hierarchical categories.
label_type = "O"
if category in entity_mapping:
label_type = entity_mapping[category]
elif category == "menu.nm":
# In the previous code it was VENDOR_NAME, but that seems wrong for menu items.
# However, to maintain compatibility with the 'ner_labels' defined in UnifiedDatasetLoader,
# we should map to what we have.
# Available: INVOICE_NUMBER, DATE, TOTAL_AMOUNT, TAX_AMOUNT, VENDOR_NAME, CUSTOMER_NAME, ADDRESS, GST_ID
# CORD is mostly food receipts.
# Let's try to find mappings.
pass
# Check for other fields manually if not in mapping
if category == "total.total_price": label_type = "TOTAL_AMOUNT"
elif category == "total.tax_price": label_type = "TAX_AMOUNT"
# For now, let's stick to a simple mapping or "O" if not sure, to avoid noise.
for word in line.get("words", []):
text = word.get("text", "")
if not text: continue
# Simple tokenization by space if needed, but usually 'text' is a word
word_tokens = text.split()
tokens.extend(word_tokens)
if label_type != "O":
labels.append(f"B-{label_type}")
labels.extend([f"I-{label_type}"] * (len(word_tokens) - 1))
else:
labels.extend(["O"] * len(word_tokens))
if tokens:
ner_data.append(
{
"tokens": tokens,
"labels": labels,
"doc_id": example.get("id", ""),
}
)
return ner_data
class SROIEDatasetLoader:
"""Load and process SROIE dataset from local archive"""
def __init__(self, base_path: str):
self.base_path = base_path
def load_dataset_splits(self):
"""Load SROIE dataset from local files"""
logger.info(f"Loading SROIE dataset from {self.base_path}...")
splits = {}
# SROIE structure: train/entities, train/box, train/img
# We will treat 'train' as train and maybe split later, or look for 'test' folder
for split_name in ["train", "test"]:
split_dir = os.path.join(self.base_path, split_name)
if not os.path.exists(split_dir):
continue
entities_dir = os.path.join(split_dir, "entities")
box_dir = os.path.join(split_dir, "box")
data = []
try:
files = os.listdir(entities_dir)
entity_files = [os.path.join(entities_dir, f) for f in files if f.endswith('.txt')]
except Exception as e:
logger.warning(f"Error listing dir {entities_dir}: {e}")
entity_files = []
for entity_file in entity_files:
file_id = os.path.basename(entity_file).replace('.txt', '')
box_file = os.path.join(box_dir, f"{file_id}.txt")
if not os.path.exists(box_file):
continue
try:
# Read entities (Ground Truth)
with open(entity_file, 'r', encoding='utf-8') as f:
# SROIE entities are usually one line JSON
entities = json.load(f)
# Read boxes and text
words = []
with open(box_file, 'r', encoding='utf-8') as f:
for line in f:
parts = line.strip().split(',')
if len(parts) >= 9:
# x1,y1,x2,y2,x3,y3,x4,y4,text
# text might contain commas, so join the rest
text = ",".join(parts[8:])
words.append(text)
data.append({
"id": file_id,
"entities": entities,
"text_lines": words,
"full_text": " ".join(words)
})
except Exception as e:
logger.warning(f"Error reading SROIE file {file_id}: {e}")
splits[split_name] = data
logger.info(f"Loaded {len(data)} examples for split '{split_name}'")
return splits
@staticmethod
def extract_classification_data(dataset_split) -> List[Dict]:
"""Extract classification data (all RECEIPT)"""
classification_data = []
for example in dataset_split:
classification_data.append(
{
"text": example.get("full_text", ""),
"label": "RECEIPT",
"image": None,
"doc_id": example.get("id", ""),
}
)
return classification_data
@staticmethod
def extract_ner_data(dataset_split) -> List[Dict]:
"""Extract NER data from SROIE"""
ner_data = []
# SROIE keys: company, date, address, total
key_mapping = {
"company": "VENDOR_NAME",
"date": "DATE",
"address": "ADDRESS",
"total": "TOTAL_AMOUNT"
}
for example in dataset_split:
text_lines = example.get("text_lines", [])
entities = example.get("entities", {})
# This is a hard problem: mapping loose entities to tokens in the text.
# For SROIE, the 'entities' file gives the *value* of the field.
# We need to find that value in the 'text_lines'.
tokens = []
labels = []
# Flatten text lines into tokens
all_tokens = []
for line in text_lines:
all_tokens.extend(line.split())
# Initialize all labels to O
token_labels = ["O"] * len(all_tokens)
# Try to match entities
# This is a naive matching approach
full_text_tokens = all_tokens
for key, value in entities.items():
if key not in key_mapping: continue
target_label = key_mapping[key]
value_tokens = value.split()
if not value_tokens: continue
# Find sequence of value_tokens in full_text_tokens
len_val = len(value_tokens)
for i in range(len(full_text_tokens) - len_val + 1):
# Check match (case insensitive? SROIE is usually exact match but OCR might vary)
# Let's try exact match first
match = True
for j in range(len_val):
if full_text_tokens[i+j] != value_tokens[j]:
match = False
break
if match:
token_labels[i] = f"B-{target_label}"
for k in range(1, len_val):
token_labels[i+k] = f"I-{target_label}"
# We only match the first occurrence for now
break
ner_data.append({
"tokens": full_text_tokens,
"labels": token_labels,
"doc_id": example.get("id", "")
})
return ner_data
class FUNSDDatasetLoader:
"""Load and process FUNSD dataset (forms) - Keeping Hugging Face for now as not in local archive"""
@staticmethod
def load_dataset_splits():
"""Load FUNSD dataset from Hugging Face"""
logger.info("Loading FUNSD dataset...")
ds = load_dataset("nielsr/funsd")
return ds
@staticmethod
def extract_classification_data(dataset_split) -> List[Dict]:
"""Extract classification data (all FORM)"""
classification_data = []
for example in dataset_split:
words = example.get("words", [])
text = " ".join(words) if words else ""
classification_data.append(
{
"text": text,
"label": "FORM",
"image": example.get("image"),
"doc_id": example.get("id", ""),
}
)
return classification_data
@staticmethod
def extract_ner_data(dataset_split) -> List[Dict]:
"""Extract NER data from FUNSD"""
ner_data = []
ner_tags_map = {
0: "O",
1: "B-HEADER",
2: "I-HEADER",
3: "B-QUESTION",
4: "I-QUESTION",
5: "B-ANSWER",
6: "I-ANSWER",
}
for example in dataset_split:
words = example.get("words", [])
ner_tags = example.get("ner_tags", [])
if not words:
continue
labels = [ner_tags_map.get(tag, "O") for tag in ner_tags]
ner_data.append(
{
"tokens": words,
"labels": labels,
"doc_id": example.get("id", ""),
}
)
return ner_data
class UnifiedDatasetLoader:
"""Unified interface for loading all datasets"""
def __init__(self):
# Define local paths
self.cord_path = r"c:/Users/Harsh-Stu/IDP[ML]/archive (2)/CORD"
self.sroie_path = r"c:/Users/Harsh-Stu/IDP[ML]/archive (1)/SROIE2019"
self.loaders = {
"cord": CORDDatasetLoader(self.cord_path),
"sroie": SROIEDatasetLoader(self.sroie_path),
"funsd": FUNSDDatasetLoader(),
}
def load_classification_dataset(
self, datasets: List[str] = ["cord"], split: str = "train"
) -> List[Dict]:
"""
Load and combine classification data from multiple datasets
"""
all_data = []
for dataset_name in datasets:
if dataset_name not in self.loaders:
logger.warning(f"Unknown dataset: {dataset_name}")
continue
try:
loader = self.loaders[dataset_name]
ds = loader.load_dataset_splits()
# Handle different split names
target_split = split
if split == "validation":
if "val" in ds: target_split = "val"
elif "dev" in ds: target_split = "dev"
elif "test" in ds: target_split = "test" # Fallback
if target_split not in ds and split == "train":
# Fallback for train if not exact match (unlikely)
pass
if target_split in ds:
data = loader.extract_classification_data(ds[target_split])
all_data.extend(data)
logger.info(
f"Loaded {len(data)} examples from {dataset_name} ({target_split})"
)
else:
logger.warning(f"Split '{target_split}' not found in {dataset_name}. Available: {list(ds.keys())}")
except Exception as e:
logger.error(f"Error loading {dataset_name}: {str(e)}")
continue
logger.info(f"Total classification examples: {len(all_data)}")
return all_data
def load_ner_dataset(
self, datasets: List[str] = ["cord", "funsd"], split: str = "train"
) -> List[Dict]:
"""
Load and combine NER data from multiple datasets
"""
all_data = []
for dataset_name in datasets:
if dataset_name not in self.loaders:
logger.warning(f"Unknown dataset: {dataset_name}")
continue
try:
loader = self.loaders[dataset_name]
ds = loader.load_dataset_splits()
# Handle different split names
target_split = split
if split == "validation":
if "val" in ds: target_split = "val"
elif "dev" in ds: target_split = "dev"
elif "test" in ds: target_split = "test"
if target_split in ds:
data = loader.extract_ner_data(ds[target_split])
all_data.extend(data)
logger.info(
f"Loaded {len(data)} NER examples from {dataset_name} ({target_split})"
)
else:
logger.warning(
f"Split '{target_split}' not found in {dataset_name}"
)
except Exception as e:
logger.error(f"Error loading {dataset_name}: {str(e)}")
continue
logger.info(f"Total NER examples: {len(all_data)}")
return all_data
def get_label_mappings(self):
"""Get label mappings for classification and NER"""
# Classification labels
classification_labels = ["INVOICE", "RECEIPT", "FORM", "OTHER"]
# NER labels (BIO tagging)
ner_labels = [
"O",
"B-INVOICE_NUMBER",
"I-INVOICE_NUMBER",
"B-DATE",
"I-DATE",
"B-TOTAL_AMOUNT",
"I-TOTAL_AMOUNT",
"B-TAX_AMOUNT",
"I-TAX_AMOUNT",
"B-VENDOR_NAME",
"I-VENDOR_NAME",
"B-CUSTOMER_NAME",
"I-CUSTOMER_NAME",
"B-ADDRESS",
"I-ADDRESS",
"B-GST_ID",
"I-GST_ID",
"B-HEADER", "I-HEADER", # FUNSD
"B-QUESTION", "I-QUESTION",
"B-ANSWER", "I-ANSWER"
]
return {
"classification": {
label: idx for idx, label in enumerate(classification_labels)
},
"ner": {label: idx for idx, label in enumerate(ner_labels)},
"classification_id2label": {
idx: label for idx, label in enumerate(classification_labels)
},
"ner_id2label": {idx: label for idx, label in enumerate(ner_labels)},
}
if __name__ == "__main__":
# Example usage
loader = UnifiedDatasetLoader()
# Load classification data
print("\n" + "=" * 50)
print("Loading Classification Data")
print("=" * 50)
classification_data = loader.load_classification_dataset(
datasets=["cord", "sroie"], split="train"
)
print(f"\nTotal examples: {len(classification_data)}")
if classification_data:
print(f"\nExample:")
print(f"Label: {classification_data[0]['label']}")
print(f"Text preview: {classification_data[0]['text'][:200]}...")
# Load NER data
print("\n" + "=" * 50)
print("Loading NER Data")
print("=" * 50)
ner_data = loader.load_ner_dataset(datasets=["cord", "sroie"], split="train")
print(f"\nTotal examples: {len(ner_data)}")
if ner_data:
print(f"\nExample:")
print(f"Tokens: {ner_data[0]['tokens'][:10]}...")
print(f"Labels: {ner_data[0]['labels'][:10]}...")
# Get label mappings
print("\n" + "=" * 50)
print("Label Mappings")
print("=" * 50)
mappings = loader.get_label_mappings()
print(f"\nClassification labels: {list(mappings['classification'].keys())}")
print(f"NER labels: {list(mappings['ner'].keys())[:10]}...")
|