Datasets:
Languages:
English
Size:
1K<n<10K
ArXiv:
Tags:
iiw
imageinwords
image-descriptions
image-captions
detailed-descriptions
hyper-detailed-descriptions
License:
roopalgarg
commited on
Commit
•
aadeaaf
1
Parent(s):
ae3ee8c
initial IIW setup
Browse files- imageinwords.py +166 -0
imageinwords.py
CHANGED
@@ -0,0 +1,166 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import datasets
|
2 |
+
import json
|
3 |
+
import pandas as pd
|
4 |
+
import string
|
5 |
+
|
6 |
+
_DESCRIPTION = """
|
7 |
+
"""
|
8 |
+
|
9 |
+
_HOMEPAGE = "https://google.github.io/imageinwords/"
|
10 |
+
|
11 |
+
_LICENSE = "CC BY 4.0"
|
12 |
+
|
13 |
+
_DATASET_GITHUB_URLS = {
|
14 |
+
"IIW-400": "https://raw.githubusercontent.com/google/imageinwords/main/datasets/IIW-400/data.jsonl?token=GHSAT0AAAAAACRUNGOO3YHB2HLA3RGXDT3SZRUODLA",
|
15 |
+
"DCI_Test": "https://raw.githubusercontent.com/google/imageinwords/main/datasets/DCI_Test/data.jsonl?token=GHSAT0AAAAAACRUNGOOB7E5WQASMIECN4JGZRUOEJQ",
|
16 |
+
"DOCCI_Test": "https://raw.githubusercontent.com/google/imageinwords/main/datasets/DOCCI_Test/data.jsonl?token=GHSAT0AAAAAACRUNGOPOYXGW4TFIX4X2LGKZRUOEVA",
|
17 |
+
"CM_3600": "https://raw.githubusercontent.com/google/imageinwords/main/datasets/CM_3600/data.jsonl?token=GHSAT0AAAAAACRUNGOO3U44BE7ICZ3LG5D2ZRUOEAA",
|
18 |
+
"LocNar_Eval": "https://raw.githubusercontent.com/google/imageinwords/main/datasets/LocNar_Eval/data.jsonl?token=GHSAT0AAAAAACRUNGOOYF254H574CK24TV2ZRUOC3Q",
|
19 |
+
}
|
20 |
+
|
21 |
+
_DATASET_FEATURES = {
|
22 |
+
"IIW-400": datasets.Features({
|
23 |
+
"image/key": datasets.Value('string'),
|
24 |
+
"image/thumbnail_url": datasets.Value('string'),
|
25 |
+
"IIW": datasets.Value('string'),
|
26 |
+
"IIW-P5B": datasets.Value('string'),
|
27 |
+
"iiw-human-sxs-gpt4v": {
|
28 |
+
"metrics/Comprehensiveness": datasets.Value('string'),
|
29 |
+
"metrics/Specificity": datasets.Value('string'),
|
30 |
+
"metrics/Hallucination": datasets.Value('string'),
|
31 |
+
"metrics/First few line(s) as tldr": datasets.Value('string'),
|
32 |
+
"metrics/Human Like": datasets.Value('string'),
|
33 |
+
},
|
34 |
+
"iiw-human-sxs-iiw-p5b": {
|
35 |
+
"metrics/Comprehensiveness": datasets.Value('string'),
|
36 |
+
"metrics/Specificity": datasets.Value('string'),
|
37 |
+
"metrics/Hallucination": datasets.Value('string'),
|
38 |
+
"metrics/First few line(s) as tldr": datasets.Value('string'),
|
39 |
+
"metrics/Human Like": datasets.Value('string'),
|
40 |
+
},
|
41 |
+
}),
|
42 |
+
"DCI_Test": datasets.Features({
|
43 |
+
"image": datasets.Value('string'),
|
44 |
+
"ex_id": datasets.Value('string'),
|
45 |
+
"IIW": datasets.Value('string'),
|
46 |
+
"metrics/Comprehensiveness": datasets.Value('string'),
|
47 |
+
"metrics/Specificity": datasets.Value('string'),
|
48 |
+
"metrics/Hallucination": datasets.Value('string'),
|
49 |
+
"metrics/First few line(s) as tldr": datasets.Value('string'),
|
50 |
+
"metrics/Human Like": datasets.Value('string'),
|
51 |
+
}),
|
52 |
+
"DOCCI_Test": datasets.Features({
|
53 |
+
"image": datasets.Value('string'),
|
54 |
+
"image/thumbnail_url": datasets.Value('string'),
|
55 |
+
"IIW": datasets.Value('string'),
|
56 |
+
"DOCCI": datasets.Value('string'),
|
57 |
+
"metrics/Comprehensiveness": datasets.Value('string'),
|
58 |
+
"metrics/Specificity": datasets.Value('string'),
|
59 |
+
"metrics/Hallucination": datasets.Value('string'),
|
60 |
+
"metrics/First few line(s) as tldr": datasets.Value('string'),
|
61 |
+
"metrics/Human Like": datasets.Value('string'),
|
62 |
+
}),
|
63 |
+
"CM_3600": datasets.Features({
|
64 |
+
"image/key": datasets.Value('string'),
|
65 |
+
"image/url": datasets.Value('string'),
|
66 |
+
"IIW-P5B": datasets.Value('string'),
|
67 |
+
}),
|
68 |
+
"LocNar_Eval": datasets.Features({
|
69 |
+
"image/key": datasets.Value('string'),
|
70 |
+
"IIW-P5B": datasets.Value('string'),
|
71 |
+
}),
|
72 |
+
}
|
73 |
+
|
74 |
+
|
75 |
+
_CM_3600_URL_PATTERN = string.Template("https://open-images-dataset.s3.amazonaws.com/crossmodal-3600/$IMAGE_KEY.jpg")
|
76 |
+
_DOCCI_AAR_THUMBNAIL_URL_PATTERN = string.Template("https://storage.googleapis.com/docci/thumbnails/$IMAGE_KEY.jpg")
|
77 |
+
|
78 |
+
|
79 |
+
class ImageInWords(datasets.GeneratorBasedBuilder):
|
80 |
+
"""ImageInWords dataset"""
|
81 |
+
|
82 |
+
VERSION = datasets.Version("1.0.0")
|
83 |
+
|
84 |
+
BUILDER_CONFIGS = [
|
85 |
+
datasets.BuilderConfig(name="IIW-400", version=VERSION, description="IIW-400"),
|
86 |
+
datasets.BuilderConfig(name="DCI_Test", version=VERSION, description="DCI_Test"),
|
87 |
+
datasets.BuilderConfig(name="DOCCI_Test", version=VERSION, description="DOCCI_Test"),
|
88 |
+
datasets.BuilderConfig(name="CM_3600", version=VERSION, description="CM_3600"),
|
89 |
+
datasets.BuilderConfig(name="LocNar_Eval", version=VERSION, description="LocNar_Eval"),
|
90 |
+
]
|
91 |
+
|
92 |
+
DEFAULT_CONFIG_NAME = "IIW-400"
|
93 |
+
|
94 |
+
def _info(self):
|
95 |
+
return datasets.DatasetInfo(
|
96 |
+
features=_DATASET_FEATURES[self.config.name],
|
97 |
+
homepage=_HOMEPAGE,
|
98 |
+
description=_DESCRIPTION,
|
99 |
+
license=_LICENSE,
|
100 |
+
)
|
101 |
+
|
102 |
+
def _split_generators(self, dl_manager):
|
103 |
+
"""Returns SplitGenerators."""
|
104 |
+
hf_auth_token = dl_manager.download_config.use_auth_token
|
105 |
+
if hf_auth_token is None:
|
106 |
+
raise ConnectionError(
|
107 |
+
"Please set use_auth_token=True or use_auth_token='<TOKEN>' to download this dataset"
|
108 |
+
)
|
109 |
+
|
110 |
+
downloaded_file = dl_manager.download_and_extract(_DATASET_GITHUB_URLS[self.config.name])
|
111 |
+
if self.config.name == "LocNar_Eval":
|
112 |
+
split_type = datasets.Split.VALIDATION
|
113 |
+
else:
|
114 |
+
split_type = datasets.Split.TEST
|
115 |
+
return [
|
116 |
+
datasets.SplitGenerator(name=split_type, gen_kwargs={"filepath": downloaded_file}),
|
117 |
+
]
|
118 |
+
|
119 |
+
def _generate_examples(self, filepath):
|
120 |
+
match self.config.name:
|
121 |
+
case "IIW-400":
|
122 |
+
return self._generate_examples_iiw_400(filepath)
|
123 |
+
case "DCI_Test":
|
124 |
+
return self._generate_examples_dci_test(filepath)
|
125 |
+
case "DOCCI_Test":
|
126 |
+
return self._generate_examples_docci_test(filepath)
|
127 |
+
case "CM_3600":
|
128 |
+
return self._generate_examples_cm_3600(filepath)
|
129 |
+
case "LocNar_Eval":
|
130 |
+
return self._generate_examples_locnar_eval(filepath)
|
131 |
+
|
132 |
+
def _generate_examples_iiw_400(filepath):
|
133 |
+
with open(filepath) as fp:
|
134 |
+
for json_line in fp:
|
135 |
+
json_obj = json.loads(json_line.strip())
|
136 |
+
json_obj["image/thumbnail_url"] = _DOCCI_AAR_THUMBNAIL_URL_PATTERN.substitute(IMAGE_KEY=json_obj["image/key"])
|
137 |
+
yield json_obj["image/key"], json_obj
|
138 |
+
|
139 |
+
def _generate_examples_dci_test(filepath):
|
140 |
+
with open(filepath) as fp:
|
141 |
+
for json_line in fp:
|
142 |
+
json_obj = json.loads(json_line.strip())
|
143 |
+
yield json_obj["image"], json_obj
|
144 |
+
|
145 |
+
def _generate_examples_docci_test(filepath):
|
146 |
+
with open(filepath) as fp:
|
147 |
+
for json_line in fp:
|
148 |
+
json_obj = json.loads(json_line.strip())
|
149 |
+
json_obj["image/thumbnail_url"] = _DOCCI_AAR_THUMBNAIL_URL_PATTERN.substitute(IMAGE_KEY=json_obj["image"])
|
150 |
+
yield json_obj["image"], json_obj
|
151 |
+
|
152 |
+
def _generate_examples_cm_3600(filepath):
|
153 |
+
with open(filepath) as fp:
|
154 |
+
for json_line in fp:
|
155 |
+
json_obj = json.loads(json_line.strip())
|
156 |
+
json_obj["image/url"] = _CM_3600_URL_PATTERN.substitute(IMAGE_KEY=json_obj["image/key"])
|
157 |
+
del json_obj["image/source"]
|
158 |
+
yield json_obj["image/key"], json_obj
|
159 |
+
|
160 |
+
|
161 |
+
def _generate_examples_locnar_eval(filepath):
|
162 |
+
with open(filepath) as fp:
|
163 |
+
for json_line in fp:
|
164 |
+
json_obj = json.loads(json_line.strip())
|
165 |
+
del json_obj["image/source"]
|
166 |
+
yield json_obj["image/key"], json_obj
|