Datasets:
wkrl
/

Sub-tasks:
parsing
Languages:
English
Multilinguality:
monolingual
Size Categories:
1K<n<10K
Language Creators:
crowdsourced
Annotations Creators:
crowdsourced
Source Datasets:
original
License:
wkrl commited on
Commit
8e8f478
1 Parent(s): 4c3a2d0

Init load script

Browse files
Files changed (1) hide show
  1. cord.py +186 -0
cord.py ADDED
@@ -0,0 +1,186 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """CORD: A Consolidated Receipt Dataset for Post-OCR Parsing"""
2
+
3
+
4
+ import json
5
+ import os
6
+ from pathlib import Path
7
+ from typing import Any, Generator
8
+
9
+ import datasets
10
+ from PIL import Image
11
+
12
+ logger = datasets.logging.get_logger(__name__)
13
+
14
+
15
+ _CITATION = """\
16
+ @article{park2019cord,
17
+ title={CORD: A Consolidated Receipt Dataset for Post-OCR Parsing},
18
+ author={Park, Seunghyun and Shin, Seung and Lee, Bado and Lee, Junyeop and Surh, Jaeheung and Seo, Minjoon and Lee, Hwalsuk}
19
+ booktitle={Document Intelligence Workshop at Neural Information Processing Systems}
20
+ year={2019}
21
+ }
22
+ """
23
+
24
+ _DESCRIPTION = """\
25
+ CORD (Consolidated Receipt Dataset) with normalized bounding boxes.
26
+ """
27
+
28
+ _URLS = [
29
+ "https://drive.google.com/uc?id=1MqhTbcj-AHXOqYoeoh12aRUwIprzTJYI",
30
+ "https://drive.google.com/uc?id=1wYdp5nC9LnHQZ2FcmOoC0eClyWvcuARU",
31
+ ]
32
+
33
+ _LABELS = [
34
+ "menu.cnt",
35
+ "menu.discountprice",
36
+ "menu.etc",
37
+ "menu.itemsubtotal",
38
+ "menu.nm",
39
+ "menu.num",
40
+ "menu.price",
41
+ "menu.sub_cnt",
42
+ "menu.sub_etc",
43
+ "menu.sub_nm",
44
+ "menu.sub_price",
45
+ "menu.sub_unitprice",
46
+ "menu.unitprice",
47
+ "menu.vatyn",
48
+ "sub_total.discount_price",
49
+ "sub_total.etc",
50
+ "sub_total.othersvc_price",
51
+ "sub_total.service_price",
52
+ "sub_total.subtotal_price",
53
+ "sub_total.tax_price",
54
+ "total.cashprice",
55
+ "total.changeprice",
56
+ "total.creditcardprice",
57
+ "total.emoneyprice",
58
+ "total.menuqty_cnt",
59
+ "total.menutype_cnt",
60
+ "total.total_etc",
61
+ "total.total_price",
62
+ "void_menu.nm",
63
+ "void_menu.price",
64
+ ]
65
+
66
+
67
+ def load_image(image_path: str) -> tuple:
68
+ image = Image.open(image_path).convert("RGB")
69
+ return image, image.size
70
+
71
+
72
+ def quad_to_bbox(quad: dict) -> list:
73
+ return [
74
+ quad["x3"],
75
+ quad["y1"],
76
+ quad["x1"],
77
+ quad["y3"],
78
+ ]
79
+
80
+
81
+ def normalize_bbox(bbox: list, width: int, height: int) -> list:
82
+ return [
83
+ int(1000 * (bbox[0] / width)),
84
+ int(1000 * (bbox[1] / height)),
85
+ int(1000 * (bbox[2] / width)),
86
+ int(1000 * (bbox[3] / height)),
87
+ ]
88
+
89
+
90
+ class CORDConfig(datasets.BuilderConfig):
91
+ """BuilderConfig for CORD."""
92
+
93
+ def __init__(self, **kwargs) -> None:
94
+ """BuilderConfig for CORD.
95
+ Args:
96
+ **kwargs: keyword arguments forwarded to super.
97
+ """
98
+ super(CORDConfig, self).__init__(**kwargs)
99
+
100
+
101
+ class CORD(datasets.GeneratorBasedBuilder):
102
+ BUILDER_CONFIGS = [
103
+ CORDConfig(
104
+ name="CORD",
105
+ version=datasets.Version("1.0.0"),
106
+ description="CORD (Consolidated Receipt Dataset)",
107
+ ),
108
+ ]
109
+
110
+ def _info(self) -> datasets.DatasetInfo:
111
+ return datasets.DatasetInfo(
112
+ description=_DESCRIPTION,
113
+ features=datasets.Features(
114
+ {
115
+ "id": datasets.Value("string"),
116
+ "words": datasets.Sequence(datasets.Value("string")),
117
+ "bboxes": datasets.Sequence(
118
+ datasets.Sequence(datasets.Value("int64"))
119
+ ),
120
+ "labels": datasets.Sequence(
121
+ datasets.features.ClassLabel(names=_LABELS)
122
+ ),
123
+ "images": datasets.features.Image(),
124
+ }
125
+ ),
126
+ citation=_CITATION,
127
+ homepage="https://github.com/clovaai/cord/",
128
+ )
129
+
130
+ def _split_generators(self, dl_manager) -> list:
131
+ base_dir_v1, base_dir_v2 = dl_manager.download_and_extract(_URLS)
132
+ dest_dir = Path(base_dir_v1) / "CORD"
133
+
134
+ for split_dir in ["train", "dev", "test"]:
135
+ for type_dir in ["image", "json"]:
136
+ if split_dir == "test" and type_dir == "json":
137
+ continue
138
+ files = (Path(base_dir_v2) / "CORD" / split_dir / type_dir).iterdir()
139
+ for f in files:
140
+ os.rename(f, dest_dir / split_dir / type_dir / f.name)
141
+
142
+ return [
143
+ datasets.SplitGenerator(
144
+ name=str(datasets.Split.TRAIN), gen_kwargs={"filepath": dest_dir / "train"}
145
+ ),
146
+ datasets.SplitGenerator(
147
+ name=str(datasets.Split.VALIDATION), gen_kwargs={"filepath": dest_dir / "dev"},
148
+ ),
149
+ datasets.SplitGenerator(
150
+ name=str(datasets.Split.TEST), gen_kwargs={"filepath": dest_dir / "test"}
151
+ ),
152
+ ]
153
+
154
+ def _generate_examples(self, **kwargs: Any) -> Generator:
155
+ filepath = kwargs["filepath"]
156
+ logger.info("generating examples from = %s", filepath)
157
+ ann_dir = os.path.join(filepath, "json")
158
+ img_dir = os.path.join(filepath, "image")
159
+
160
+ for guid, file in enumerate(sorted(os.listdir(ann_dir))):
161
+ WORDS, BBOXES, LABELS = [], [], []
162
+ file_path = os.path.join(ann_dir, file)
163
+ f = open(file_path)
164
+ data = json.load(f)
165
+
166
+ image_path = os.path.join(img_dir, file).replace("json", "png")
167
+ image, (width, height) = load_image(image_path)
168
+
169
+ for annotation in data["valid_line"]:
170
+ label, words = annotation["category"], annotation["words"]
171
+ for word in words:
172
+ bbox = normalize_bbox(
173
+ quad_to_bbox(word["quad"]), width=width, height=height
174
+ )
175
+ if min(bbox) >= 0 and max(bbox) <= 1000:
176
+ WORDS.append(word["text"])
177
+ BBOXES.append(bbox)
178
+ LABELS.append(label)
179
+
180
+ yield guid, {
181
+ "id": str(guid),
182
+ "images": image,
183
+ "words": WORDS,
184
+ "bboxes": BBOXES,
185
+ "labels": LABELS,
186
+ }