Vadzim Kashko commited on
Commit
39559a6
1 Parent(s): d9f44d5

feat: add load script

Browse files
Files changed (1) hide show
  1. ocr-barcodes-detection.py +157 -0
ocr-barcodes-detection.py ADDED
@@ -0,0 +1,157 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from xml.etree import ElementTree as ET
2
+
3
+ import datasets
4
+
5
+ _CITATION = """\
6
+ @InProceedings{huggingface:dataset,
7
+ title = {ocr-barcodes-detection},
8
+ author = {TrainingDataPro},
9
+ year = {2023}
10
+ }
11
+ """
12
+
13
+ _DESCRIPTION = """\
14
+ The Grocery Store Receipts Dataset is a collection of photos captured from various
15
+ **grocery store receipts**. This dataset is specifically designed for tasks related to
16
+ **Optical Character Recognition (OCR)** and is useful for retail.
17
+ Each image in the dataset is accompanied by bounding box annotations, indicating the
18
+ precise locations of specific text segments on the receipts. The text segments are
19
+ categorized into four classes: **item, store, date_time and total**.
20
+ """
21
+ _NAME = "ocr-barcodes-detection"
22
+
23
+ _HOMEPAGE = f"https://huggingface.co/datasets/TrainingDataPro/{_NAME}"
24
+
25
+ _LICENSE = ""
26
+
27
+ _DATA = f"https://huggingface.co/datasets/TrainingDataPro/{_NAME}/resolve/main/data/"
28
+
29
+ _LABELS = ["Barcode"]
30
+
31
+
32
+ class OcrBarcodesDetection(datasets.GeneratorBasedBuilder):
33
+ def _info(self):
34
+ return datasets.DatasetInfo(
35
+ description=_DESCRIPTION,
36
+ features=datasets.Features(
37
+ {
38
+ "id": datasets.Value("int32"),
39
+ "name": datasets.Value("string"),
40
+ "image": datasets.Image(),
41
+ "mask": datasets.Image(),
42
+ "width": datasets.Value("uint16"),
43
+ "height": datasets.Value("uint16"),
44
+ "shapes": datasets.Sequence(
45
+ {
46
+ "label": datasets.ClassLabel(
47
+ num_classes=len(_LABELS),
48
+ names=_LABELS,
49
+ ),
50
+ "type": datasets.Value("string"),
51
+ "points": datasets.Sequence(
52
+ datasets.Sequence(
53
+ datasets.Value("float"),
54
+ ),
55
+ ),
56
+ "rotation": datasets.Value("float"),
57
+ "occluded": datasets.Value("uint8"),
58
+ "attributes": datasets.Sequence(
59
+ {
60
+ "name": datasets.Value("string"),
61
+ "text": datasets.Value("string"),
62
+ }
63
+ ),
64
+ }
65
+ ),
66
+ }
67
+ ),
68
+ supervised_keys=None,
69
+ homepage=_HOMEPAGE,
70
+ citation=_CITATION,
71
+ )
72
+
73
+ def _split_generators(self, dl_manager):
74
+ images = dl_manager.download(f"{_DATA}images.tar.gz")
75
+ masks = dl_manager.download(f"{_DATA}boxes.tar.gz")
76
+ annotations = dl_manager.download(f"{_DATA}annotations.xml")
77
+ images = dl_manager.iter_archive(images)
78
+ masks = dl_manager.iter_archive(masks)
79
+ return [
80
+ datasets.SplitGenerator(
81
+ name=datasets.Split.TRAIN,
82
+ gen_kwargs={
83
+ "images": images,
84
+ "masks": masks,
85
+ "annotations": annotations,
86
+ },
87
+ ),
88
+ ]
89
+
90
+ @staticmethod
91
+ def parse_shape(shape: ET.Element) -> dict:
92
+ label = shape.get("label")
93
+ shape_type = shape.tag
94
+ rotation = shape.get("rotation", 0.0)
95
+ occluded = shape.get("occluded", 0)
96
+
97
+ points = None
98
+
99
+ if shape_type == "points":
100
+ points = tuple(map(float, shape.get("points").split(",")))
101
+
102
+ elif shape_type == "box":
103
+ points = [
104
+ (float(shape.get("xtl")), float(shape.get("ytl"))),
105
+ (float(shape.get("xbr")), float(shape.get("ybr"))),
106
+ ]
107
+
108
+ elif shape_type == "polygon":
109
+ points = [
110
+ tuple(map(float, point.split(",")))
111
+ for point in shape.get("points").split(";")
112
+ ]
113
+
114
+ attributes = []
115
+
116
+ for attr in shape:
117
+ attr_name = attr.get("name")
118
+ attr_text = attr.text
119
+ attributes.append({"name": attr_name, "text": attr_text})
120
+
121
+ shape_data = {
122
+ "label": label,
123
+ "type": shape_type,
124
+ "points": points,
125
+ "rotation": rotation,
126
+ "occluded": occluded,
127
+ "attributes": attributes,
128
+ }
129
+
130
+ return shape_data
131
+
132
+ def _generate_examples(self, images, masks, annotations):
133
+ tree = ET.parse(annotations)
134
+ root = tree.getroot()
135
+
136
+ for idx, (
137
+ (image_path, image),
138
+ (mask_path, mask),
139
+ ) in enumerate(zip(images, masks)):
140
+ image_name = image_path.split("/")[-1]
141
+ img = root.find(f"./image[@name='images/{image_name}']")
142
+
143
+ image_id = img.get("id")
144
+ name = img.get("name")
145
+ width = img.get("width")
146
+ height = img.get("height")
147
+ shapes = [self.parse_shape(shape) for shape in img]
148
+
149
+ yield idx, {
150
+ "id": image_id,
151
+ "name": name,
152
+ "image": {"path": image_path, "bytes": image.read()},
153
+ "mask": {"path": mask_path, "bytes": mask.read()},
154
+ "width": width,
155
+ "height": height,
156
+ "shapes": shapes,
157
+ }