Datasets:

Languages:
Thai
ArXiv:
License:
holylovenia commited on
Commit
262da91
1 Parent(s): 760f6c2

Upload alice_thi.py with huggingface_hub

Browse files
Files changed (1) hide show
  1. alice_thi.py +261 -0
alice_thi.py ADDED
@@ -0,0 +1,261 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # coding=utf-8
2
+ # Copyright 2022 The HuggingFace Datasets Authors and the current dataset script contributor.
3
+ #
4
+ # Licensed under the Apache License, Version 2.0 (the "License");
5
+ # you may not use this file except in compliance with the License.
6
+ # You may obtain a copy of the License at
7
+ #
8
+ # http://www.apache.org/licenses/LICENSE-2.0
9
+ #
10
+ # Unless required by applicable law or agreed to in writing, software
11
+ # distributed under the License is distributed on an "AS IS" BASIS,
12
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13
+ # See the License for the specific language governing permissions and
14
+ # limitations under the License.
15
+
16
+ from pathlib import Path
17
+ from typing import Dict, List, Tuple
18
+
19
+ import datasets
20
+
21
+ from seacrowd.utils import schemas
22
+ from seacrowd.utils.configs import SEACrowdConfig
23
+ from seacrowd.utils.constants import TASK_TO_SCHEMA, Licenses, Tasks
24
+
25
+ _CITATION = """\
26
+ @article{SURINTA2015405,
27
+ title = "Recognition of handwritten characters using local gradient feature descriptors",
28
+ journal = "Engineering Applications of Artificial Intelligence",
29
+ volume = "45",
30
+ number = "Supplement C",
31
+ pages = "405 - 414",
32
+ year = "2015",
33
+ issn = "0952-1976",
34
+ doi = "https://doi.org/10.1016/j.engappai.2015.07.017",
35
+ url = "http://www.sciencedirect.com/science/article/pii/S0952197615001724",
36
+ author = "Olarik Surinta and Mahir F. Karaaba and Lambert R.B. Schomaker and Marco A. Wiering",
37
+ keywords = "Handwritten character recognition, Feature extraction, Local gradient feature descriptor,
38
+ Support vector machine, k-nearest neighbors"
39
+ }
40
+ """
41
+
42
+ _DATASETNAME = "alice_thi"
43
+
44
+ _DESCRIPTION = """\
45
+ ALICE-THI is a Thai handwritten script dataset that contains 24045 character
46
+ images, which is split into Thai handwritten character dataset (THI-C68) for
47
+ 14490 images and Thai handwritten digit dataset (THI-D10) for 9555 images. The
48
+ data was collected from 150 native writers aged from 20 to 23 years old. The
49
+ participants were allowed to write only the isolated Thai script on the form and
50
+ at least 100 samples per character. The character images obtained from this
51
+ dataset generally have no background noise.
52
+ """
53
+
54
+ _HOMEPAGE = "https://www.ai.rug.nl/~mrolarik/ALICE-THI/"
55
+
56
+ _LANGUAGES = ["tha"]
57
+ _SUBSETS = {
58
+ "THI-D10": {
59
+ "data_dir": "Thai_digit_sqr",
60
+ "label_dict": {
61
+ 0: "0",
62
+ 1: "1",
63
+ 2: "2",
64
+ 3: "3",
65
+ 4: "4",
66
+ 5: "5",
67
+ 6: "6",
68
+ 7: "7",
69
+ 8: "8",
70
+ 9: "9",
71
+ },
72
+ },
73
+ "THI-C68": {
74
+ "data_dir": "Thai_char_sqr",
75
+ "label_dict": {
76
+ 0: "ko kai",
77
+ 1: "kho khai",
78
+ 2: "kho khuat",
79
+ 3: "kho khwai",
80
+ 4: "kho khon",
81
+ 5: "kho rakhang",
82
+ 6: "ngo ngu",
83
+ 7: "cho chan",
84
+ 8: "cho ching",
85
+ 9: "cho chang",
86
+ 10: "so so",
87
+ 11: "cho choe",
88
+ 12: "yo ying",
89
+ 13: "do chada",
90
+ 14: "to patak",
91
+ 15: "tho than",
92
+ 16: "tho nangmontho",
93
+ 17: "tho phuthao",
94
+ 18: "no nen",
95
+ 19: "do dek",
96
+ 20: "to tao",
97
+ 21: "tho thung",
98
+ 22: "tho thahan",
99
+ 23: "tho thong",
100
+ 24: "no nu",
101
+ 25: "bo baimai",
102
+ 26: "po pla",
103
+ 27: "pho phung",
104
+ 28: "fo fa",
105
+ 29: "pho phan",
106
+ 30: "fo fan",
107
+ 31: "pho samphao",
108
+ 32: "mo ma",
109
+ 33: "yo yak",
110
+ 34: "ro rua",
111
+ 35: "ru",
112
+ 36: "lo ling",
113
+ 37: "lu",
114
+ 38: "wo waen",
115
+ 39: "so rusi",
116
+ 40: "so sala",
117
+ 41: "so sua",
118
+ 42: "ho hip",
119
+ 43: "lo chula",
120
+ 44: "o ang",
121
+ 45: "ho nokhuk",
122
+ 46: "paiyannoi",
123
+ 47: "sara a",
124
+ 48: "mai han",
125
+ 49: "sara aa",
126
+ 50: "sara i",
127
+ 51: "sara ii",
128
+ 52: "sara ue",
129
+ 53: "sara uee",
130
+ 54: "sara u",
131
+ 55: "sara uu",
132
+ 56: "sara e",
133
+ 57: "sara o",
134
+ 58: "sara ai maimuan",
135
+ 59: "sara ai maimalai",
136
+ 60: "maiyamok",
137
+ 61: "maitaikhu",
138
+ 62: "mai ek",
139
+ 63: "mai tho",
140
+ 64: "mai tri",
141
+ 65: "mai chattawa",
142
+ 66: "thanthakhat",
143
+ 67: "nikhahit",
144
+ },
145
+ },
146
+ }
147
+
148
+ _LICENSE = Licenses.UNKNOWN.value
149
+
150
+ _LOCAL = False
151
+
152
+ _URLS = {
153
+ _DATASETNAME: "https://www.ai.rug.nl/~mrolarik/ALICE-THI/ALICE-THI-Dataset.tar.gz",
154
+ }
155
+
156
+ _SUPPORTED_TASKS = [Tasks.OPTICAL_CHARACTER_RECOGNITION]
157
+ _SEACROWD_SCHEMA = f"seacrowd_{TASK_TO_SCHEMA[_SUPPORTED_TASKS[0]].lower()}" # imtext
158
+
159
+ _SOURCE_VERSION = "1.0.0"
160
+
161
+ _SEACROWD_VERSION = "2024.06.20"
162
+
163
+
164
+ class AliceTHIDataset(datasets.GeneratorBasedBuilder):
165
+ """Thai handwritten script dataset for character and digit recognition."""
166
+
167
+ SOURCE_VERSION = datasets.Version(_SOURCE_VERSION)
168
+ SEACROWD_VERSION = datasets.Version(_SEACROWD_VERSION)
169
+
170
+ BUILDER_CONFIGS = []
171
+ for subset in list(_SUBSETS.keys()):
172
+ BUILDER_CONFIGS += [
173
+ SEACrowdConfig(
174
+ name=f"{_DATASETNAME}_{subset}_source",
175
+ version=SOURCE_VERSION,
176
+ description=f"{_DATASETNAME} {subset} source schema",
177
+ schema="source",
178
+ subset_id=subset,
179
+ ),
180
+ SEACrowdConfig(
181
+ name=f"{_DATASETNAME}_{subset}_{_SEACROWD_SCHEMA}",
182
+ version=SEACROWD_VERSION,
183
+ description=f"{_DATASETNAME} {subset} SEACrowd schema",
184
+ schema=_SEACROWD_SCHEMA,
185
+ subset_id=subset,
186
+ ),
187
+ ]
188
+
189
+ DEFAULT_CONFIG_NAME = f"{_DATASETNAME}_THI-C68_source"
190
+
191
+ def _info(self) -> datasets.DatasetInfo:
192
+ label_names = [val for _, val in sorted(_SUBSETS[self.config.subset_id]["label_dict"].items())]
193
+ if self.config.schema == "source":
194
+ features = datasets.Features(
195
+ {
196
+ "label": datasets.ClassLabel(names=label_names),
197
+ "text": datasets.Value("string"),
198
+ "image_path": datasets.Value("string"),
199
+ }
200
+ )
201
+ elif self.config.schema == _SEACROWD_SCHEMA:
202
+ features = schemas.image_text_features(label_names=label_names)
203
+
204
+ return datasets.DatasetInfo(
205
+ description=_DESCRIPTION,
206
+ features=features,
207
+ homepage=_HOMEPAGE,
208
+ license=_LICENSE,
209
+ citation=_CITATION,
210
+ )
211
+
212
+ def _split_generators(self, dl_manager: datasets.DownloadManager) -> List[datasets.SplitGenerator]:
213
+ """Returns SplitGenerators."""
214
+ data_name = "ALICE-THI Dataset"
215
+ data_path = Path(dl_manager.download_and_extract(_URLS[_DATASETNAME]))
216
+ data_path = Path(dl_manager.extract(data_path / data_name / f"{data_name}.tar.gz"))
217
+ data_path = data_path / _SUBSETS[self.config.subset_id]["data_dir"]
218
+
219
+ return [
220
+ datasets.SplitGenerator(
221
+ name=datasets.Split.TRAIN,
222
+ gen_kwargs={
223
+ "data_path": data_path,
224
+ },
225
+ ),
226
+ ]
227
+
228
+ def _generate_examples(self, data_path: Path) -> Tuple[int, Dict]:
229
+ """Yields examples as (key, example) tuples."""
230
+ # iterate over files and directories
231
+ for subfolder in data_path.iterdir():
232
+ if subfolder.is_dir():
233
+
234
+ # source schema yield one image per label
235
+ if self.config.schema == "source":
236
+ _get_label = True # efficiency placeholder
237
+ for image_file in subfolder.glob("*.png"):
238
+ if _get_label: # get label from filename
239
+ label = int(image_file.name.split("-")[0].lower())
240
+ _get_label = False
241
+
242
+ yield image_file.stem, {
243
+ "label": label,
244
+ "text": _SUBSETS[self.config.subset_id]["label_dict"][label],
245
+ "image_path": str(image_file),
246
+ }
247
+
248
+ # seacrowd schema yield multiple images per label
249
+ elif self.config.schema == _SEACROWD_SCHEMA:
250
+ image_files = list(subfolder.glob("*.png"))
251
+ label = int(image_files[0].name.split("-")[0].lower())
252
+
253
+ yield subfolder.name, {
254
+ "id": subfolder.name,
255
+ "image_paths": [str(file) for file in image_files],
256
+ "texts": _SUBSETS[self.config.subset_id]["label_dict"][label],
257
+ "metadata": {
258
+ "context": "",
259
+ "labels": [label] * len(image_files),
260
+ },
261
+ }