holylovenia commited on
Commit
b7475cd
1 Parent(s): 51a8a52

Upload vivqa.py with huggingface_hub

Browse files
Files changed (1) hide show
  1. vivqa.py +218 -0
vivqa.py ADDED
@@ -0,0 +1,218 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # coding=utf-8
2
+ import json
3
+ import os
4
+ from pathlib import Path
5
+ from typing import Dict, List, Tuple
6
+
7
+ import datasets
8
+ import pandas as pd
9
+
10
+ from seacrowd.utils import schemas
11
+ from seacrowd.utils.configs import SEACrowdConfig
12
+ from seacrowd.utils.constants import Licenses, Tasks
13
+
14
+ _CITATION = """\
15
+ @inproceedings{tran2021vivqa,
16
+ title={ViVQA: Vietnamese visual question answering},
17
+ author={Tran, Khanh Quoc and Nguyen, An Trong and Le, An Tran-Hoai and Van Nguyen, Kiet},
18
+ booktitle={Proceedings of the 35th Pacific Asia Conference on Language, Information and Computation},
19
+ pages={683--691},
20
+ year={2021}
21
+ }
22
+ """
23
+ _DATASETNAME = "vivqa"
24
+ _DESCRIPTION = """\
25
+ Vietnamese Visual Question Answering (ViVQA) consist of 10328 images and 15000 question-answer
26
+ pairs in Vietnamese for evaluating Vietnamese VQA models. This dataset is built based on 10328 randomly
27
+ selected images from MS COCO dataset. The question-answer pairs were based on the COCO-QA dataset that
28
+ was automatically translated from English to Vietnamese.
29
+ """
30
+ _HOMEPAGE = "https://github.com/kh4nh12/ViVQA"
31
+ _LANGUAGES = ["vie"]
32
+ _LICENSE = Licenses.UNKNOWN.value
33
+ _LOCAL = False
34
+ _URLS = {
35
+ "viviq": {"train": "https://raw.githubusercontent.com/kh4nh12/ViVQA/main/train.csv",
36
+ "test": "https://raw.githubusercontent.com/kh4nh12/ViVQA/main/test.csv"},
37
+ "cocodata": {
38
+ "coco2014_train_val_annots": "http://images.cocodataset.org/annotations/annotations_trainval2014.zip",
39
+ "coco2014_train_images": "http://images.cocodataset.org/zips/train2014.zip",
40
+ "coco2014_val_images": "http://images.cocodataset.org/zips/val2014.zip",
41
+ },
42
+ }
43
+ _SUPPORTED_TASKS = [Tasks.VISUAL_QUESTION_ANSWERING]
44
+ _SOURCE_VERSION = "1.0.0"
45
+ _SEACROWD_VERSION = "2024.06.20"
46
+
47
+
48
+ class VivQADataset(datasets.GeneratorBasedBuilder):
49
+ SOURCE_VERSION = datasets.Version(_SOURCE_VERSION)
50
+ SEACROWD_VERSION = datasets.Version(_SEACROWD_VERSION)
51
+
52
+ BUILDER_CONFIGS = [
53
+ SEACrowdConfig(
54
+ name=f"{_DATASETNAME}_source",
55
+ version=SOURCE_VERSION,
56
+ description=f"{_DATASETNAME} source schema",
57
+ schema="source",
58
+ subset_id=f"{_DATASETNAME}",
59
+ ),
60
+ SEACrowdConfig(
61
+ name=f"{_DATASETNAME}_seacrowd_imqa",
62
+ version=SEACROWD_VERSION,
63
+ description=f"{_DATASETNAME} SEACrowd schema",
64
+ schema="seacrowd_imqa",
65
+ subset_id=f"{_DATASETNAME}",
66
+ ),
67
+ ]
68
+
69
+ DEFAULT_CONFIG_NAME = f"{_DATASETNAME}_source"
70
+
71
+ def _info(self) -> datasets.DatasetInfo:
72
+
73
+ if self.config.schema == "source":
74
+ features = datasets.Features(
75
+ {
76
+ "img_id": datasets.Value("string"),
77
+ "question": datasets.Value("string"),
78
+ "answer": datasets.Value("string"),
79
+ "type": datasets.Value("string"),
80
+ "coco_url": datasets.Value("string"),
81
+ "flickr_url": datasets.Value("string"),
82
+ "img_name": datasets.Value("string"),
83
+ "coco_license": datasets.Value("int32"),
84
+ "coco_width": datasets.Value("int32"),
85
+ "coco_height": datasets.Value("int32"),
86
+ "coco_date_captured": datasets.Value("string"),
87
+ "image_path": datasets.Value("string"),
88
+ }
89
+ )
90
+ elif self.config.schema == "seacrowd_imqa":
91
+ features = schemas.imqa_features
92
+ features["meta"] = {
93
+ "coco_img_id": datasets.Value("string"),
94
+ "type": datasets.Value("string"),
95
+ "flickr_url": datasets.Value("string"),
96
+ "coco_url": datasets.Value("string"),
97
+ "img_name": datasets.Value("string"),
98
+ "coco_license": datasets.Value("int32"),
99
+ "coco_width": datasets.Value("int32"),
100
+ "coco_height": datasets.Value("int32"),
101
+ "coco_date_captured": datasets.Value("string"),
102
+ "image_path": datasets.Value("string"),
103
+ }
104
+ else:
105
+ raise ValueError(f"No schema matched for {self.config.schema}")
106
+
107
+ return datasets.DatasetInfo(
108
+ description=_DESCRIPTION,
109
+ features=features,
110
+ homepage=_HOMEPAGE,
111
+ license=_LICENSE,
112
+ citation=_CITATION,
113
+ )
114
+
115
+ def _split_generators(self, dl_manager: datasets.DownloadManager) -> List[datasets.SplitGenerator]:
116
+ """Returns SplitGenerators."""
117
+ urls = _URLS["viviq"]
118
+ data_dir = dl_manager.download_and_extract(urls)
119
+ cocodata = dl_manager.download_and_extract(_URLS["cocodata"])
120
+ Coco_Dict = self._get_image_detail(cocodata)
121
+
122
+ return [
123
+ datasets.SplitGenerator(
124
+ name=datasets.Split.TRAIN,
125
+ gen_kwargs={
126
+ "filepath": data_dir["train"],
127
+ "split": "train",
128
+ "coco_dict": Coco_Dict,
129
+ },
130
+ ),
131
+ datasets.SplitGenerator(
132
+ name=datasets.Split.TEST,
133
+ gen_kwargs={
134
+ "filepath": data_dir["test"],
135
+ "split": "test",
136
+ "coco_dict": Coco_Dict,
137
+ },
138
+ ),
139
+ ]
140
+
141
+ def _get_image_detail(self, coco_dir) -> Dict:
142
+ coco2014_train_val_annots = os.path.join(coco_dir["coco2014_train_val_annots"], "annotations")
143
+ train_ann_2014_path = os.path.join(coco2014_train_val_annots, "captions_train2014.json")
144
+ val_ann_2014_path = os.path.join(coco2014_train_val_annots, "captions_val2014.json")
145
+ coco_dict_val = {itm["id"]: itm for itm in json.load(open(val_ann_2014_path, "r"))["images"]}
146
+ coco_dict_train = {itm["id"]: itm for itm in json.load(open(train_ann_2014_path, "r"))["images"]}
147
+ coco_train_path = os.path.join(coco_dir["coco2014_train_images"], "train2014")
148
+ coco_val_path = os.path.join(coco_dir["coco2014_val_images"], "val2014")
149
+ coco_dict = {"train": coco_dict_train, "val": coco_dict_val, "coco_train_path": coco_train_path, "coco_val_path": coco_val_path}
150
+
151
+ return coco_dict
152
+
153
+ def _generate_examples(self, filepath: Path, split: str, coco_dict: Dict = None) -> Tuple[int, Dict]:
154
+ """Yields examples as (key, example) tuples."""
155
+
156
+ raw_examples = pd.read_csv(filepath)
157
+ coco_train_ref = coco_dict["train"]
158
+ coco_val_ref = coco_dict["val"]
159
+ coco_ref = {**coco_train_ref, **coco_val_ref}
160
+ coco_train_path = coco_dict["coco_train_path"]
161
+ coco_val_path = coco_dict["coco_val_path"]
162
+
163
+ for eid, exam in raw_examples.iterrows():
164
+ assert len(exam) == 5
165
+ exam_id, exam_quest, exam_answer, exam_img_id, exam_type = exam
166
+ coco_info = coco_ref[exam_img_id]
167
+ flickr_url = coco_info["flickr_url"]
168
+ img_name = coco_info["file_name"]
169
+ coco_url = coco_info["coco_url"]
170
+ coco_license = coco_info["license"]
171
+ coco_width = coco_info["width"]
172
+ coco_height = coco_info["height"]
173
+ coco_date_captured = coco_info["date_captured"]
174
+ coco_path = coco_train_path if exam_img_id in coco_train_ref else coco_val_path
175
+ image_path = os.path.join(coco_path, img_name)
176
+
177
+ if self.config.schema == "source":
178
+ yield eid, {
179
+ "img_id": str(exam_img_id),
180
+ "question": exam_quest,
181
+ "answer": exam_answer,
182
+ "type": exam_type,
183
+ "coco_url": coco_url,
184
+ "flickr_url": flickr_url,
185
+ "img_name": img_name,
186
+ "coco_license": coco_license,
187
+ "coco_width": coco_width,
188
+ "coco_height": coco_height,
189
+ "coco_date_captured": coco_date_captured,
190
+ "image_path": image_path,
191
+ }
192
+
193
+ elif self.config.schema == "seacrowd_imqa":
194
+ example = {
195
+ "id": str(eid),
196
+ "question_id": str(exam_id),
197
+ "document_id": str(eid),
198
+ "questions": [exam_quest],
199
+ "type": None,
200
+ "choices": None,
201
+ "context": None,
202
+ "answer": [exam_answer],
203
+ "image_paths": [image_path],
204
+ "meta": {
205
+ "coco_img_id": str(exam_img_id),
206
+ "type": exam_type,
207
+ "flickr_url": flickr_url,
208
+ "coco_url": coco_url,
209
+ "img_name": img_name,
210
+ "coco_license": coco_license,
211
+ "coco_width": coco_width,
212
+ "coco_height": coco_height,
213
+ "coco_date_captured": coco_date_captured,
214
+ "image_path": image_path,
215
+ },
216
+ }
217
+
218
+ yield eid, example