nielsr HF staff commited on
Commit
5227407
1 Parent(s): e88572e

Create funsd-layoutlmv3.py

Browse files
Files changed (1) hide show
  1. funsd-layoutlmv3.py +132 -0
funsd-layoutlmv3.py ADDED
@@ -0,0 +1,132 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # coding=utf-8
2
+ '''
3
+ Reference: https://huggingface.co/datasets/nielsr/funsd/blob/main/funsd.py
4
+ '''
5
+ import json
6
+ import os
7
+
8
+ import datasets
9
+
10
+ from layoutlmft.data.image_utils import load_image, normalize_bbox
11
+
12
+
13
+ logger = datasets.logging.get_logger(__name__)
14
+
15
+
16
+ _CITATION = """\
17
+ @article{Jaume2019FUNSDAD,
18
+ title={FUNSD: A Dataset for Form Understanding in Noisy Scanned Documents},
19
+ author={Guillaume Jaume and H. K. Ekenel and J. Thiran},
20
+ journal={2019 International Conference on Document Analysis and Recognition Workshops (ICDARW)},
21
+ year={2019},
22
+ volume={2},
23
+ pages={1-6}
24
+ }
25
+ """
26
+
27
+ _DESCRIPTION = """\
28
+ https://guillaumejaume.github.io/FUNSD/
29
+ """
30
+
31
+
32
+ class FunsdConfig(datasets.BuilderConfig):
33
+ """BuilderConfig for FUNSD"""
34
+
35
+ def __init__(self, **kwargs):
36
+ """BuilderConfig for FUNSD.
37
+
38
+ Args:
39
+ **kwargs: keyword arguments forwarded to super.
40
+ """
41
+ super(FunsdConfig, self).__init__(**kwargs)
42
+
43
+
44
+ class Funsd(datasets.GeneratorBasedBuilder):
45
+ """Conll2003 dataset."""
46
+
47
+ BUILDER_CONFIGS = [
48
+ FunsdConfig(name="funsd", version=datasets.Version("1.0.0"), description="FUNSD dataset"),
49
+ ]
50
+
51
+ def _info(self):
52
+ return datasets.DatasetInfo(
53
+ description=_DESCRIPTION,
54
+ features=datasets.Features(
55
+ {
56
+ "id": datasets.Value("string"),
57
+ "tokens": datasets.Sequence(datasets.Value("string")),
58
+ "bboxes": datasets.Sequence(datasets.Sequence(datasets.Value("int64"))),
59
+ "ner_tags": datasets.Sequence(
60
+ datasets.features.ClassLabel(
61
+ names=["O", "B-HEADER", "I-HEADER", "B-QUESTION", "I-QUESTION", "B-ANSWER", "I-ANSWER"]
62
+ )
63
+ ),
64
+ "image": datasets.Array3D(shape=(3, 224, 224), dtype="uint8"),
65
+ "image_path": datasets.Value("string"),
66
+ }
67
+ ),
68
+ supervised_keys=None,
69
+ homepage="https://guillaumejaume.github.io/FUNSD/",
70
+ citation=_CITATION,
71
+ )
72
+
73
+ def _split_generators(self, dl_manager):
74
+ """Returns SplitGenerators."""
75
+ downloaded_file = dl_manager.download_and_extract("https://guillaumejaume.github.io/FUNSD/dataset.zip")
76
+ return [
77
+ datasets.SplitGenerator(
78
+ name=datasets.Split.TRAIN, gen_kwargs={"filepath": f"{downloaded_file}/dataset/training_data/"}
79
+ ),
80
+ datasets.SplitGenerator(
81
+ name=datasets.Split.TEST, gen_kwargs={"filepath": f"{downloaded_file}/dataset/testing_data/"}
82
+ ),
83
+ ]
84
+
85
+ def get_line_bbox(self, bboxs):
86
+ x = [bboxs[i][j] for i in range(len(bboxs)) for j in range(0, len(bboxs[i]), 2)]
87
+ y = [bboxs[i][j] for i in range(len(bboxs)) for j in range(1, len(bboxs[i]), 2)]
88
+
89
+ x0, y0, x1, y1 = min(x), min(y), max(x), max(y)
90
+
91
+ assert x1 >= x0 and y1 >= y0
92
+ bbox = [[x0, y0, x1, y1] for _ in range(len(bboxs))]
93
+ return bbox
94
+
95
+ def _generate_examples(self, filepath):
96
+ logger.info("⏳ Generating examples from = %s", filepath)
97
+ ann_dir = os.path.join(filepath, "annotations")
98
+ img_dir = os.path.join(filepath, "images")
99
+ for guid, file in enumerate(sorted(os.listdir(ann_dir))):
100
+ tokens = []
101
+ bboxes = []
102
+ ner_tags = []
103
+
104
+ file_path = os.path.join(ann_dir, file)
105
+ with open(file_path, "r", encoding="utf8") as f:
106
+ data = json.load(f)
107
+ image_path = os.path.join(img_dir, file)
108
+ image_path = image_path.replace("json", "png")
109
+ image, size = load_image(image_path)
110
+ for item in data["form"]:
111
+ cur_line_bboxes = []
112
+ words, label = item["words"], item["label"]
113
+ words = [w for w in words if w["text"].strip() != ""]
114
+ if len(words) == 0:
115
+ continue
116
+ if label == "other":
117
+ for w in words:
118
+ tokens.append(w["text"])
119
+ ner_tags.append("O")
120
+ cur_line_bboxes.append(normalize_bbox(w["box"], size))
121
+ else:
122
+ tokens.append(words[0]["text"])
123
+ ner_tags.append("B-" + label.upper())
124
+ cur_line_bboxes.append(normalize_bbox(words[0]["box"], size))
125
+ for w in words[1:]:
126
+ tokens.append(w["text"])
127
+ ner_tags.append("I-" + label.upper())
128
+ cur_line_bboxes.append(normalize_bbox(w["box"], size))
129
+ cur_line_bboxes = self.get_line_bbox(cur_line_bboxes)
130
+ bboxes.extend(cur_line_bboxes)
131
+ yield guid, {"id": str(guid), "tokens": tokens, "bboxes": bboxes, "ner_tags": ner_tags,
132
+ "image": image, "image_path": image_path}