Datasets:
Tasks:
Visual Question Answering
Formats:
parquet
Languages:
English
Size:
10K - 100K
ArXiv:
Tags:
medical
License:
Commit
·
c3944be
1
Parent(s):
83b3c8e
Upload path_vqa.py
Browse files- path_vqa.py +135 -0
path_vqa.py
ADDED
@@ -0,0 +1,135 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""PathVQA: 30000+ Questions for Medical Visual Question Answering"""
|
2 |
+
|
3 |
+
import pandas
|
4 |
+
import os
|
5 |
+
|
6 |
+
import datasets
|
7 |
+
|
8 |
+
_CITATION = """\
|
9 |
+
@article{he2020pathvqa,
|
10 |
+
title={PathVQA: 30000+ Questions for Medical Visual Question Answering},
|
11 |
+
author={He, Xuehai and Zhang, Yichen and Mou, Luntian and Xing, Eric and Xie, Pengtao},
|
12 |
+
journal={arXiv preprint arXiv:2003.10286},
|
13 |
+
year={2020}
|
14 |
+
}
|
15 |
+
"""
|
16 |
+
|
17 |
+
_DESCRIPTION = """\
|
18 |
+
PathVQA is a dataset of question-answer pairs on pathology images. The questions are similar to those in the
|
19 |
+
American Board of Pathology (ABP) test. The dataset includes both open-ended questions and binary "yes/no"
|
20 |
+
questions. The dataset is built from two publicly-available pathology textbooks: "Textbook of Pathology" and
|
21 |
+
"Basic Pathology", and a publicly-available digital library: "Pathology Education Informational Resource"
|
22 |
+
(PEIR). The copyrights of images and captions belong to the publishers and authors of these two books,
|
23 |
+
and the owners of the PEIR digital library.
|
24 |
+
"""
|
25 |
+
|
26 |
+
_HOMEPAGE = "https://github.com/UCSD-AI4H/PathVQA"
|
27 |
+
|
28 |
+
_LICENSE = "MIT"
|
29 |
+
|
30 |
+
_URLS = {
|
31 |
+
"image_train": "data/image/train_img.tar",
|
32 |
+
"image_val": "data/image/val_img.tar",
|
33 |
+
"image_test": "data/image/test_img.tar",
|
34 |
+
"text_train": "data/text/train_qa.jsonl",
|
35 |
+
"text_val": "data/text/val_qa.jsonl",
|
36 |
+
"text_test": "data/text/test_qa.jsonl",
|
37 |
+
}
|
38 |
+
|
39 |
+
class PathVQA(datasets.GeneratorBasedBuilder):
|
40 |
+
|
41 |
+
"""
|
42 |
+
PathVQA: 30000+ Questions for Medical Visual Question Answering.
|
43 |
+
|
44 |
+
The data was obtained from the updated Google Drive link shared by the authors in their GitHub repository
|
45 |
+
on Feb 15, 2023, see https://github.com/UCSD-AI4H/PathVQA/commit/117e7f4ef88a0e65b0e7f37b98a73d6237a3ceab.
|
46 |
+
|
47 |
+
This version of the dataset contains a total of 5,004 images and 32,795 question-answer pairs. Of the
|
48 |
+
5,004 images, 4,289 images are referenced by a question-answer pair, while 715 images are not used.
|
49 |
+
Furthermore, there are several duplicates, i.e. there are some image-question-answer triplets which occur
|
50 |
+
more than once in the same split (train, val, test). After dropping the duplicate image-question-answer
|
51 |
+
triplets, the dataset contains 32,632 question-answer pairs on 4,289 images.
|
52 |
+
"""
|
53 |
+
|
54 |
+
VERSION = datasets.Version("0.1.0")
|
55 |
+
|
56 |
+
BUILDER_CONFIGS = [
|
57 |
+
datasets.BuilderConfig(name="full", version=VERSION, description="Original dataset."),
|
58 |
+
datasets.BuilderConfig(name="de-duped", version=VERSION, description="De-duplicated dataset."),
|
59 |
+
]
|
60 |
+
|
61 |
+
DEFAULT_CONFIG_NAME = "de-duped"
|
62 |
+
|
63 |
+
def _info(self):
|
64 |
+
|
65 |
+
features = datasets.Features(
|
66 |
+
{
|
67 |
+
"image": datasets.Image(),
|
68 |
+
"question": datasets.Value("string"),
|
69 |
+
"answer": datasets.Value("string")
|
70 |
+
}
|
71 |
+
)
|
72 |
+
|
73 |
+
return datasets.DatasetInfo(
|
74 |
+
description=_DESCRIPTION,
|
75 |
+
features=features,
|
76 |
+
homepage=_HOMEPAGE,
|
77 |
+
license=_LICENSE,
|
78 |
+
citation=_CITATION,
|
79 |
+
)
|
80 |
+
|
81 |
+
def _split_generators(self, dl_manager):
|
82 |
+
|
83 |
+
# images
|
84 |
+
image_train_dir = dl_manager.download_and_extract(_URLS["image_train"])
|
85 |
+
image_val_dir = dl_manager.download_and_extract(_URLS["image_val"])
|
86 |
+
image_test_dir = dl_manager.download_and_extract(_URLS["image_test"])
|
87 |
+
|
88 |
+
# question-answer pairs
|
89 |
+
text_train_dir = dl_manager.download(_URLS["text_train"])
|
90 |
+
text_val_dir = dl_manager.download(_URLS["text_val"])
|
91 |
+
text_test_dir = dl_manager.download(_URLS["text_test"])
|
92 |
+
|
93 |
+
return [
|
94 |
+
|
95 |
+
datasets.SplitGenerator(
|
96 |
+
name=datasets.Split.TRAIN,
|
97 |
+
gen_kwargs={
|
98 |
+
"image_filepath": os.path.join(image_train_dir),
|
99 |
+
"text_filepath": os.path.join(text_train_dir),
|
100 |
+
"split": "train",
|
101 |
+
},
|
102 |
+
),
|
103 |
+
|
104 |
+
datasets.SplitGenerator(
|
105 |
+
name=datasets.Split.VALIDATION,
|
106 |
+
gen_kwargs={
|
107 |
+
"image_filepath": os.path.join(image_val_dir),
|
108 |
+
"text_filepath": os.path.join(text_val_dir),
|
109 |
+
"split": "val",
|
110 |
+
},
|
111 |
+
),
|
112 |
+
|
113 |
+
datasets.SplitGenerator(
|
114 |
+
name=datasets.Split.TEST,
|
115 |
+
gen_kwargs={
|
116 |
+
"image_filepath": os.path.join(image_test_dir),
|
117 |
+
"text_filepath": os.path.join(text_test_dir),
|
118 |
+
"split": "test"
|
119 |
+
},
|
120 |
+
),
|
121 |
+
]
|
122 |
+
|
123 |
+
def _generate_examples(self, image_filepath, text_filepath, split):
|
124 |
+
|
125 |
+
df = pandas.read_json(text_filepath, orient='records', lines=True)
|
126 |
+
if self.config.name == "de-duped":
|
127 |
+
df = df.drop_duplicates(ignore_index=True)
|
128 |
+
|
129 |
+
for key, row in df.iterrows():
|
130 |
+
yield key, {
|
131 |
+
"image": os.path.join(image_filepath, row['image']),
|
132 |
+
"question": row["question"],
|
133 |
+
"answer": row["answer"]
|
134 |
+
}
|
135 |
+
|