Create sample.py
Browse files
sample.py
ADDED
@@ -0,0 +1,77 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import csv
|
2 |
+
|
3 |
+
import datasets
|
4 |
+
from datasets import DatasetDict
|
5 |
+
|
6 |
+
LABELS = {"aerial", "interior", "exterior", "upshot", "skyline", "night"}
|
7 |
+
|
8 |
+
_DATA_URL = {
|
9 |
+
"train": [f"data/train_images.tar.gz" for i in range(5)],
|
10 |
+
"validation": ["data/validation_images.tar.gz"],
|
11 |
+
"test": ["data/test_images.tar.gz"],
|
12 |
+
}
|
13 |
+
|
14 |
+
|
15 |
+
class Sample(datasets.GeneratorBasedBuilder):
|
16 |
+
DEFAULT_WRITER_BATCH_SIZE = 1000
|
17 |
+
|
18 |
+
def _info(self):
|
19 |
+
return datasets.DatasetInfo(
|
20 |
+
description="A sample dataset to illustrate how to use HF APIs",
|
21 |
+
features=datasets.Features(
|
22 |
+
{
|
23 |
+
"image": datasets.Image(),
|
24 |
+
"label": datasets.ClassLabel(names=list(LABELS)),
|
25 |
+
}
|
26 |
+
),
|
27 |
+
homepage="github.com/SOM-Enterprise/hf-dataset-sample-representation",
|
28 |
+
citation="None",
|
29 |
+
task_templates=[
|
30 |
+
datasets.ImageClassification(image_column="image", label_column="label")],
|
31 |
+
)
|
32 |
+
|
33 |
+
def _split_generators(self, dl_manager):
|
34 |
+
archives = dl_manager.download(_DATA_URL)
|
35 |
+
return [
|
36 |
+
datasets.SplitGenerator(
|
37 |
+
name=datasets.Split.TRAIN,
|
38 |
+
gen_kwargs={
|
39 |
+
"archives": [dl_manager.iter_archive(archive) for archive in archives["train"]],
|
40 |
+
"split": "train",
|
41 |
+
}
|
42 |
+
),
|
43 |
+
datasets.SplitGenerator(
|
44 |
+
name=datasets.Split.VALIDATION,
|
45 |
+
gen_kwargs={
|
46 |
+
"archives": [dl_manager.iter_archive(archive) for archive in archives["validation"]],
|
47 |
+
"split": "validation",
|
48 |
+
}
|
49 |
+
),
|
50 |
+
datasets.SplitGenerator(
|
51 |
+
name=datasets.Split.TEST,
|
52 |
+
gen_kwargs={
|
53 |
+
"archives": [dl_manager.iter_archive(archive) for archive in archives["test"]],
|
54 |
+
"split": "test",
|
55 |
+
}
|
56 |
+
)
|
57 |
+
]
|
58 |
+
|
59 |
+
def _generate_examples(self, archives, split):
|
60 |
+
labels_dict = {}
|
61 |
+
with open('metadata.csv', newline='') as csvfile:
|
62 |
+
reader = csv.DictReader(csvfile)
|
63 |
+
for row in reader:
|
64 |
+
labels_dict[row['id']] = set(row['label'].split('|'))
|
65 |
+
|
66 |
+
idx = 0
|
67 |
+
for archive in archives:
|
68 |
+
for path, file in archive:
|
69 |
+
if path.endswith(".jpeg"):
|
70 |
+
if split != "test":
|
71 |
+
labels = labels_dict.get(path.split('/')[-1])
|
72 |
+
label = labels if labels else ['']
|
73 |
+
else:
|
74 |
+
label = -1
|
75 |
+
ex = {"image": {"path": path, "bytes": file.read()}, "label": label}
|
76 |
+
yield idx, ex
|
77 |
+
idx += 1
|