jamescalam
commited on
Commit
•
61bf1ad
1
Parent(s):
2d7603c
Create new file
Browse files- image-text-demo.py +60 -0
image-text-demo.py
ADDED
@@ -0,0 +1,60 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import datasets
|
2 |
+
|
3 |
+
_CITATION = """\
|
4 |
+
@InProceedings{huggingface:dataset,
|
5 |
+
title = {Small image-text set},
|
6 |
+
author={James Briggs},
|
7 |
+
year={2022}
|
8 |
+
}
|
9 |
+
"""
|
10 |
+
|
11 |
+
_DESCRIPTION = """\
|
12 |
+
Demo dataset for testing or showing image-text capabilities.
|
13 |
+
"""
|
14 |
+
_HOMEPAGE = "https://huggingface.co/datasets/jamescalam/image-set"
|
15 |
+
|
16 |
+
_LICENSE = ""
|
17 |
+
|
18 |
+
_REPO = "https://huggingface.co/datasets/jamescalam/image-set"
|
19 |
+
|
20 |
+
class ImageSet(datasets.GeneratorBasedBuilder):
|
21 |
+
"""Small sample of image-text pairs"""
|
22 |
+
|
23 |
+
def _info(self):
|
24 |
+
return datasets.DatasetInfo(
|
25 |
+
description=_DESCRIPTION,
|
26 |
+
features=datasets.Features(
|
27 |
+
{
|
28 |
+
'id': datasets.Value("int32"),
|
29 |
+
'text': datasets.Value("string"),
|
30 |
+
'image': datasets.Image(),
|
31 |
+
}
|
32 |
+
),
|
33 |
+
supervised_keys=None,
|
34 |
+
homepage=_HOMEPAGE,
|
35 |
+
citation=_CITATION,
|
36 |
+
)
|
37 |
+
|
38 |
+
def _split_generators(self, dl_manager):
|
39 |
+
images_path = dl_manager.download_and_extract(f"{_REPO}/resolve/main/unsplash.tgz")
|
40 |
+
image_iters = dl_manager.iter_archive(f"{images_path}/images")
|
41 |
+
return [
|
42 |
+
datasets.SplitGenerator(
|
43 |
+
name=datasets.Split.TRAIN,
|
44 |
+
gen_kwargs={
|
45 |
+
"images": image_iters
|
46 |
+
}
|
47 |
+
),
|
48 |
+
]
|
49 |
+
|
50 |
+
def _generate_examples(self, images):
|
51 |
+
"""This function returns the examples in the raw (text) form."""
|
52 |
+
id_ = 0
|
53 |
+
for filepath, image in images:
|
54 |
+
description = filepath.split('/')[-1][:-4]
|
55 |
+
description = description.replace('_', ' ')
|
56 |
+
yield id_, {
|
57 |
+
"image": {"path": filepath, "bytes": image.read()},
|
58 |
+
"text": description,
|
59 |
+
}
|
60 |
+
idx += 1
|