echarlaix HF staff commited on
Commit
33b7d67
1 Parent(s): cb9871f

Commit file to HF hub

Browse files
Files changed (1) hide show
  1. gqa.py +99 -0
gqa.py ADDED
@@ -0,0 +1,99 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # coding=utf-8
2
+ # Copyright 2020 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
+ """The GQA dataset."""
16
+
17
+ import json
18
+ import os
19
+
20
+ import datasets
21
+
22
+
23
+ _CITATION = """\
24
+ @inproceedings{hudson2019gqa,
25
+ title={Gqa: A new dataset for real-world visual reasoning and compositional question answering},
26
+ author={Hudson, Drew A and Manning, Christopher D},
27
+ booktitle={Proceedings of the IEEE/CVF conference on computer vision and pattern recognition},
28
+ pages={6700--6709},
29
+ year={2019}
30
+ }
31
+ """
32
+
33
+ _DESCRIPTION = """\
34
+ GQA is a new dataset for real-world visual reasoning and compositional question answering,
35
+ seeking to address key shortcomings of previous visual question answering (VQA) datasets.
36
+ """
37
+
38
+ _URLS = {
39
+ "train": "https://nlp.cs.unc.edu/data/lxmert_data/gqa/train.json",
40
+ "dev": "https://nlp.cs.unc.edu/data/lxmert_data/gqa/valid.json",
41
+ "img": "https://downloads.cs.stanford.edu/nlp/data/gqa/images.zip",
42
+ "ans2label": "https://raw.githubusercontent.com/airsplay/lxmert/master/data/gqa/trainval_ans2label.json",
43
+ }
44
+
45
+ _IMG_DIR = "images"
46
+
47
+
48
+ class Gqa(datasets.GeneratorBasedBuilder):
49
+ """The GQA dataset."""
50
+
51
+ BUILDER_CONFIGS = [
52
+ datasets.BuilderConfig(name="gqa", version=datasets.Version("1.0.0"), description="GQA dataset."),
53
+ ]
54
+
55
+ def _info(self):
56
+ features = datasets.Features(
57
+ {
58
+ "input_ids": datasets.Value("string"),
59
+ "img_id": datasets.Value("string"),
60
+ "question_id": datasets.Value("int32"),
61
+ "label": datasets.Value("int32"),
62
+ }
63
+ )
64
+ return datasets.DatasetInfo(
65
+ description=_DESCRIPTION,
66
+ features=features,
67
+ supervised_keys=None,
68
+ citation=_CITATION,
69
+ )
70
+
71
+ def _split_generators(self, dl_manager):
72
+ """Returns SplitGenerators."""
73
+ dl_dir = dl_manager.download_and_extract(_URLS)
74
+ self.ans2label = json.load(open(dl_dir["ans2label"]))
75
+
76
+ return [
77
+ datasets.SplitGenerator(
78
+ name=datasets.Split.TRAIN,
79
+ gen_kwargs={"filepath": dl_dir["train"], "img_dir": os.path.join(dl_dir["img"], _IMG_DIR)},
80
+ ),
81
+ datasets.SplitGenerator(
82
+ name=datasets.Split.VALIDATION,
83
+ gen_kwargs={"filepath": dl_dir["dev"], "img_dir": os.path.join(dl_dir["img"], _IMG_DIR)},
84
+ ),
85
+ ]
86
+
87
+ def _generate_examples(self, filepath, img_dir):
88
+ """ Yields examples as (key, example) tuples. """
89
+ with open(filepath, encoding="utf-8") as f:
90
+ gqa = json.load(f)
91
+ for id_, d in enumerate(gqa):
92
+ img_id = os.path.join(img_dir, d["img_id"] + ".jpg")
93
+ label = self.ans2label[next(iter(d["label"].keys()))]
94
+ yield id_, {
95
+ "input_ids": d["sent"],
96
+ "img_id": img_id,
97
+ "label": label,
98
+ "question_id": d["question_id"],
99
+ }