Fixed issue with test file 12052d3b8a19248565c63470ef4b3088 - changed ImDB loaders
Browse files- DUDE_imdb_loader.py +20 -29
DUDE_imdb_loader.py
CHANGED
@@ -77,9 +77,7 @@ def pdf_to_images(document_filepath, converter="PyPDF2"):
|
|
77 |
def images_to_pagenames(images, document_filepath, page_image_dir):
|
78 |
page_image_names = []
|
79 |
for page_idx, page_image in enumerate(images):
|
80 |
-
page_image_name = document_filepath.replace("PDF", "images").replace(
|
81 |
-
".pdf", f"_{page_idx}.jpg"
|
82 |
-
)
|
83 |
page_image_names.append(
|
84 |
page_image_name.replace(page_image_dir, page_image_dir.split("/")[-1])
|
85 |
) # without dir
|
@@ -172,7 +170,7 @@ def get_ocr_information(ocr_path, num_pages):
|
|
172 |
|
173 |
ocr_pages = ocr_info[0]["DocumentMetadata"]["Pages"]
|
174 |
|
175 |
-
if num_pages != ocr_pages:
|
176 |
raise AssertionError("Pages from images and OCR not matching, should go for pdf2image")
|
177 |
|
178 |
page_ocr_tokens = [[] for page_ix in range(num_pages)]
|
@@ -181,11 +179,12 @@ def get_ocr_information(ocr_path, num_pages):
|
|
181 |
for ocr_extraction in ocr_block["Blocks"]:
|
182 |
if ocr_extraction["BlockType"] == "WORD":
|
183 |
text = ocr_extraction["Text"].lower()
|
184 |
-
bounding_box = parse_textract_bbox(
|
185 |
-
ocr_extraction["Geometry"]["BoundingBox"]
|
186 |
-
).tolist()
|
187 |
page = ocr_extraction["Page"] - 1
|
188 |
|
|
|
|
|
|
|
189 |
page_ocr_tokens[page].append(text)
|
190 |
page_ocr_boxes[page].append(bounding_box)
|
191 |
|
@@ -222,12 +221,14 @@ def format_answers(answers_list):
|
|
222 |
def create_imdb_record_from_json(
|
223 |
record, documents_metadata, documents_ocr_info, split, include_answers, include_variants=False
|
224 |
):
|
225 |
-
|
226 |
docId = record["docId"].split("_")[0]
|
227 |
try:
|
228 |
num_pages, page_image_names = get_document_info(documents_metadata, docId)
|
229 |
document_ocr_info = documents_ocr_info[docId]
|
230 |
except Exception as e:
|
|
|
|
|
|
|
231 |
print(
|
232 |
"Missing: ",
|
233 |
e,
|
@@ -240,17 +241,15 @@ def create_imdb_record_from_json(
|
|
240 |
else:
|
241 |
answers = None
|
242 |
|
243 |
-
if include_variants and record["answers_variants"] and not
|
244 |
answers += record["answers_variants"]
|
245 |
|
246 |
page_image_dir = "/".join(record["document"].split("/")[:-2]).replace("PDF", "images")
|
247 |
-
if not page_image_names or any(
|
248 |
-
[not os.path.exists(os.path.join(page_image_dir, p)) for p in page_image_names]
|
249 |
-
):
|
250 |
print(
|
251 |
"Missing images: ",
|
252 |
docId,
|
253 |
-
#[p for p in page_image_names if not os.path.exists(os.path.join(page_image_dir, p))],
|
254 |
)
|
255 |
return {}
|
256 |
|
@@ -308,9 +307,7 @@ def create_imdb_from_json(
|
|
308 |
def parse_arguments():
|
309 |
import argparse
|
310 |
|
311 |
-
parser = argparse.ArgumentParser(
|
312 |
-
description="Instantiate HuggingFace dataloader and convert to ImDB format"
|
313 |
-
)
|
314 |
|
315 |
parser.add_argument(
|
316 |
"--redo-imdb-build",
|
@@ -379,17 +376,13 @@ if __name__ == "__main__":
|
|
379 |
num_jobs = 6
|
380 |
block_size = int(len(document_paths) / num_jobs) + 1
|
381 |
print(f"{block_size} * {num_jobs} = {block_size*num_jobs} ({len(document_paths)})")
|
382 |
-
document_blocks = [
|
383 |
-
document_paths[block_size * i : block_size * i + block_size]
|
384 |
-
for i in range(num_jobs)
|
385 |
-
]
|
386 |
print(
|
387 |
"chunksize",
|
388 |
len(set([docId for doc_block in document_blocks for docId in doc_block])),
|
389 |
)
|
390 |
parallel_results = Parallel(n_jobs=num_jobs)(
|
391 |
-
delayed(pdf_to_images_block)(document_blocks[i], "pdf2image")
|
392 |
-
for i in range(num_jobs)
|
393 |
)
|
394 |
|
395 |
for block_result in parallel_results:
|
@@ -413,9 +406,7 @@ if __name__ == "__main__":
|
|
413 |
for i, document_filepath in enumerate(document_paths):
|
414 |
docId = document_filepath.split("/")[-1].replace(".pdf", "")
|
415 |
try:
|
416 |
-
ocr_tokens, ocr_boxes = get_ocr_information(
|
417 |
-
OCR_paths[i], documents_metadata[docId]["num_pages"]
|
418 |
-
)
|
419 |
documents_ocr_info[docId] = {"ocr_tokens": ocr_tokens, "ocr_boxes": ocr_boxes}
|
420 |
except AssertionError as e:
|
421 |
print(f"image2pages issue: {e}")
|
@@ -423,10 +414,10 @@ if __name__ == "__main__":
|
|
423 |
except IndexError as e:
|
424 |
print(f"pages issue: {e}")
|
425 |
error_ocr.append(docId)
|
426 |
-
except FileNotFoundError:
|
427 |
print(f"FileNotFoundError issue: {e}")
|
428 |
no_ocr.append(docId)
|
429 |
-
except KeyError:
|
430 |
print(f"Keyerror issue: {e}")
|
431 |
error_ocr.append(docId)
|
432 |
|
@@ -437,14 +428,14 @@ if __name__ == "__main__":
|
|
437 |
print(f"Loading from disk: {imdb_filename}")
|
438 |
imdb = np.load(imdb_filename, allow_pickle=True)
|
439 |
|
440 |
-
else:
|
441 |
imdb = create_imdb_from_json(
|
442 |
dataset[split], # .select(split_indices),
|
443 |
documents_metadata=documents_metadata,
|
444 |
documents_ocr_info=documents_ocr_info,
|
445 |
split=split,
|
446 |
version="0.1",
|
447 |
-
include_answers=
|
448 |
include_variants=(not args.no_include_variants),
|
449 |
)
|
450 |
np.save(imdb_filename, imdb)
|
|
|
77 |
def images_to_pagenames(images, document_filepath, page_image_dir):
|
78 |
page_image_names = []
|
79 |
for page_idx, page_image in enumerate(images):
|
80 |
+
page_image_name = document_filepath.replace("PDF", "images").replace(".pdf", f"_{page_idx}.jpg")
|
|
|
|
|
81 |
page_image_names.append(
|
82 |
page_image_name.replace(page_image_dir, page_image_dir.split("/")[-1])
|
83 |
) # without dir
|
|
|
170 |
|
171 |
ocr_pages = ocr_info[0]["DocumentMetadata"]["Pages"]
|
172 |
|
173 |
+
if num_pages != ocr_pages and num_pages != MAX_PAGES: # MAX_PAGES is the limit for conversion
|
174 |
raise AssertionError("Pages from images and OCR not matching, should go for pdf2image")
|
175 |
|
176 |
page_ocr_tokens = [[] for page_ix in range(num_pages)]
|
|
|
179 |
for ocr_extraction in ocr_block["Blocks"]:
|
180 |
if ocr_extraction["BlockType"] == "WORD":
|
181 |
text = ocr_extraction["Text"].lower()
|
182 |
+
bounding_box = parse_textract_bbox(ocr_extraction["Geometry"]["BoundingBox"]).tolist()
|
|
|
|
|
183 |
page = ocr_extraction["Page"] - 1
|
184 |
|
185 |
+
if page >= num_pages: # additional condition when MAX_PAGES vs. OCR pages
|
186 |
+
break
|
187 |
+
|
188 |
page_ocr_tokens[page].append(text)
|
189 |
page_ocr_boxes[page].append(bounding_box)
|
190 |
|
|
|
221 |
def create_imdb_record_from_json(
|
222 |
record, documents_metadata, documents_ocr_info, split, include_answers, include_variants=False
|
223 |
):
|
|
|
224 |
docId = record["docId"].split("_")[0]
|
225 |
try:
|
226 |
num_pages, page_image_names = get_document_info(documents_metadata, docId)
|
227 |
document_ocr_info = documents_ocr_info[docId]
|
228 |
except Exception as e:
|
229 |
+
from pdb import set_trace
|
230 |
+
|
231 |
+
set_trace()
|
232 |
print(
|
233 |
"Missing: ",
|
234 |
e,
|
|
|
241 |
else:
|
242 |
answers = None
|
243 |
|
244 |
+
if include_variants and record["answers_variants"] and not "list" in record["answer_type"]:
|
245 |
answers += record["answers_variants"]
|
246 |
|
247 |
page_image_dir = "/".join(record["document"].split("/")[:-2]).replace("PDF", "images")
|
248 |
+
if not page_image_names or any([not os.path.exists(os.path.join(page_image_dir, p)) for p in page_image_names]):
|
|
|
|
|
249 |
print(
|
250 |
"Missing images: ",
|
251 |
docId,
|
252 |
+
# [p for p in page_image_names if not os.path.exists(os.path.join(page_image_dir, p))],
|
253 |
)
|
254 |
return {}
|
255 |
|
|
|
307 |
def parse_arguments():
|
308 |
import argparse
|
309 |
|
310 |
+
parser = argparse.ArgumentParser(description="Instantiate HuggingFace dataloader and convert to ImDB format")
|
|
|
|
|
311 |
|
312 |
parser.add_argument(
|
313 |
"--redo-imdb-build",
|
|
|
376 |
num_jobs = 6
|
377 |
block_size = int(len(document_paths) / num_jobs) + 1
|
378 |
print(f"{block_size} * {num_jobs} = {block_size*num_jobs} ({len(document_paths)})")
|
379 |
+
document_blocks = [document_paths[block_size * i : block_size * i + block_size] for i in range(num_jobs)]
|
|
|
|
|
|
|
380 |
print(
|
381 |
"chunksize",
|
382 |
len(set([docId for doc_block in document_blocks for docId in doc_block])),
|
383 |
)
|
384 |
parallel_results = Parallel(n_jobs=num_jobs)(
|
385 |
+
delayed(pdf_to_images_block)(document_blocks[i], "pdf2image") for i in range(num_jobs)
|
|
|
386 |
)
|
387 |
|
388 |
for block_result in parallel_results:
|
|
|
406 |
for i, document_filepath in enumerate(document_paths):
|
407 |
docId = document_filepath.split("/")[-1].replace(".pdf", "")
|
408 |
try:
|
409 |
+
ocr_tokens, ocr_boxes = get_ocr_information(OCR_paths[i], documents_metadata[docId]["num_pages"])
|
|
|
|
|
410 |
documents_ocr_info[docId] = {"ocr_tokens": ocr_tokens, "ocr_boxes": ocr_boxes}
|
411 |
except AssertionError as e:
|
412 |
print(f"image2pages issue: {e}")
|
|
|
414 |
except IndexError as e:
|
415 |
print(f"pages issue: {e}")
|
416 |
error_ocr.append(docId)
|
417 |
+
except FileNotFoundError as e:
|
418 |
print(f"FileNotFoundError issue: {e}")
|
419 |
no_ocr.append(docId)
|
420 |
+
except KeyError as e:
|
421 |
print(f"Keyerror issue: {e}")
|
422 |
error_ocr.append(docId)
|
423 |
|
|
|
428 |
print(f"Loading from disk: {imdb_filename}")
|
429 |
imdb = np.load(imdb_filename, allow_pickle=True)
|
430 |
|
431 |
+
else:
|
432 |
imdb = create_imdb_from_json(
|
433 |
dataset[split], # .select(split_indices),
|
434 |
documents_metadata=documents_metadata,
|
435 |
documents_ocr_info=documents_ocr_info,
|
436 |
split=split,
|
437 |
version="0.1",
|
438 |
+
include_answers=(not split == "test"),
|
439 |
include_variants=(not args.no_include_variants),
|
440 |
)
|
441 |
np.save(imdb_filename, imdb)
|