Datasets:
Size:
1M<n<10M
ArXiv:
Tags:
Document_Understanding
Document_Packet_Splitting
Document_Comprehension
Document_Classification
Document_Recognition
Document_Segmentation
DOI:
License:
File size: 7,716 Bytes
165da3c | 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 | # Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
# SPDX-License-Identifier: CC-BY-NC-4.0
import csv
import json
from pathlib import Path
from loguru import logger
from models import BenchmarkSet
class BenchmarkWriter:
"""Writes benchmark datasets to disk as CSV files."""
def __init__(self, output_base_path: str, assets_path: str = "data/assets"):
self.output_base_path = Path(output_base_path)
self.assets_path = Path(assets_path)
def save_benchmark_set(
self,
benchmark_set: BenchmarkSet,
split: str
):
"""Save a benchmark set to CSV and ground truth JSON files.
Args:
benchmark_set: The benchmark set to save
split: Split name (train, test, or validation)
"""
self.output_base_path.mkdir(parents=True, exist_ok=True)
output_path = self.output_base_path / f"{split}.csv"
json_dir = self.output_base_path / "ground_truth_json" / split
json_dir.mkdir(parents=True, exist_ok=True)
# Generate group IDs (AA, AB, AC, etc.)
def generate_group_id(index):
"""Generate group_id like AA, AB, AC, ..., AZ, BA, BB, etc."""
alphabet = "ABCDEFGHIJKLMNOPQRSTUVWXYZ"
first_char = alphabet[index // 26 % 26]
second_char = alphabet[index % 26]
return first_char + second_char
# Flatten ground truth pages into CSV rows
rows = []
# Track unique source documents across all packets for group_id assignment
source_doc_to_group_id = {}
group_id_counter = 0
for doc_idx, doc in enumerate(benchmark_set.documents):
# Track page ordinal per source document
source_doc_ordinals = {}
# Track local_doc_id per source document within this packet
source_doc_to_local_id = {}
local_id_counter = {}
for gt_page in doc.ground_truth:
source_key = gt_page.source_doc
# Assign group_id per unique source document
if source_key not in source_doc_to_group_id:
source_doc_to_group_id[source_key] = generate_group_id(group_id_counter)
group_id_counter += 1
group_id = source_doc_to_group_id[source_key]
# Increment ordinal for this source document
if source_key not in source_doc_ordinals:
source_doc_ordinals[source_key] = 0
source_doc_ordinals[source_key] += 1
# Assign local_doc_id per source document
if source_key not in source_doc_to_local_id:
doc_type_prefix = gt_page.doc_type.replace(" ", "-").lower()
if doc_type_prefix not in local_id_counter:
local_id_counter[doc_type_prefix] = 0
local_id_counter[doc_type_prefix] += 1
source_doc_to_local_id[source_key] = f"{doc_type_prefix}-{local_id_counter[doc_type_prefix]:02d}"
local_doc_id = source_doc_to_local_id[source_key]
# Build paths
image_path = f"{self.assets_path}/{gt_page.doc_type}/{gt_page.source_doc}.pdf/pages/{gt_page.source_page:04d}/page-{gt_page.source_page:04d}.png"
text_path = f"{self.assets_path}/{gt_page.doc_type}/{gt_page.source_doc}.pdf/pages/{gt_page.source_page:04d}/page-{gt_page.source_page:04d}-textract.md"
rows.append({
'doc_type': gt_page.doc_type,
'original_doc_name': gt_page.source_doc,
'parent_doc_name': doc.spliced_doc_id,
'local_doc_id': local_doc_id,
'page': gt_page.page_num,
'image_path': image_path,
'text_path': text_path,
'group_id': group_id,
'local_doc_id_page_ordinal': source_doc_ordinals[source_key]
})
# Generate ground truth JSON for this parent document
self._save_ground_truth_json(doc, json_dir)
# Write CSV
if rows:
fieldnames = ['doc_type', 'original_doc_name', 'parent_doc_name', 'local_doc_id',
'page', 'image_path', 'text_path', 'group_id', 'local_doc_id_page_ordinal']
with open(output_path, 'w', newline='', encoding='utf-8') as f:
writer = csv.DictWriter(f, fieldnames=fieldnames)
writer.writeheader()
writer.writerows(rows)
logger.info(f"Saved {len(benchmark_set.documents)} spliced documents ({len(rows)} pages) to {output_path}")
logger.info(f"Saved {len(benchmark_set.documents)} ground truth JSON files to {json_dir}")
def _save_ground_truth_json(self, doc, json_dir: Path):
"""Generate ground truth JSON for a single parent document."""
# Group pages by source document (doc_type, source_doc)
subdocs = {}
for gt_page in doc.ground_truth:
key = (gt_page.doc_type, gt_page.source_doc)
if key not in subdocs:
subdocs[key] = []
subdocs[key].append(gt_page)
# Build subdocuments structure
subdocuments = []
for subdoc_idx, ((doc_type, source_doc), pages) in enumerate(subdocs.items()):
group_id = self._generate_group_id(subdoc_idx)
# Count occurrences of this doc_type so far
doc_type_count = sum(1 for s in subdocuments if s['doc_type_id'] == doc_type) + 1
local_doc_id = f"{doc_type}-{doc_type_count:02d}"
page_ordinals = [p.page_num for p in pages]
subdoc_pages = []
for idx, gt_page in enumerate(pages, 1):
image_path = f"{self.assets_path}/{gt_page.doc_type}/{gt_page.source_doc}.pdf/pages/{gt_page.source_page:04d}/page-{gt_page.source_page:04d}.png"
text_path = f"{self.assets_path}/{gt_page.doc_type}/{gt_page.source_doc}.pdf/pages/{gt_page.source_page:04d}/page-{gt_page.source_page:04d}-textract.md"
subdoc_pages.append({
"page": gt_page.page_num,
"original_doc_name": gt_page.source_doc,
"image_path": image_path,
"text_path": text_path,
"local_doc_id_page_ordinal": idx
})
subdocuments.append({
"doc_type_id": doc_type,
"page_ordinals": page_ordinals,
"local_doc_id": local_doc_id,
"group_id": group_id,
"pages": subdoc_pages
})
# Build final JSON structure
ground_truth = {
"doc_id": doc.spliced_doc_id,
"total_pages": doc.total_pages,
"subdocuments": subdocuments
}
# Write JSON file
json_path = json_dir / f"{doc.spliced_doc_id}.json"
with open(json_path, 'w', encoding='utf-8') as f:
json.dump(ground_truth, f, indent=2)
def _generate_group_id(self, index):
"""Generate group_id like AA, AB, AC, ..., AZ, BA, BB, etc."""
result = ""
index += 1
while index > 0:
index -= 1
result = chr(65 + (index % 26)) + result
index //= 26
return result or "A"
|