Create medium.py
Browse files
medium.py
ADDED
@@ -0,0 +1,100 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import datasets
|
2 |
+
import json
|
3 |
+
import numpy
|
4 |
+
import tarfile
|
5 |
+
import io
|
6 |
+
|
7 |
+
_FEATURES = datasets.Features(
|
8 |
+
{
|
9 |
+
"id": datasets.Value("string"),
|
10 |
+
"prompt": datasets.Array3D(shape=(1, 77, 768), dtype="float32"),
|
11 |
+
"video": datasets.Sequence(feature=datasets.Array3D(shape=(4, 64, 64), dtype="float32")),
|
12 |
+
"description": datasets.Value("string"),
|
13 |
+
"videourl": datasets.Value("string"),
|
14 |
+
"categories": datasets.Value("string"),
|
15 |
+
"duration": datasets.Value("float"),
|
16 |
+
"full_metadata": datasets.Value("string"),
|
17 |
+
}
|
18 |
+
)
|
19 |
+
|
20 |
+
class FunkLoaderStream(datasets.GeneratorBasedBuilder):
|
21 |
+
"""TempoFunk Dataset"""
|
22 |
+
|
23 |
+
def _info(self):
|
24 |
+
return datasets.DatasetInfo(
|
25 |
+
description="TempoFunk Dataset",
|
26 |
+
features=_FEATURES,
|
27 |
+
homepage="tempofunk.github.io",
|
28 |
+
citation="""
|
29 |
+
@misc{TempoFunk2023,
|
30 |
+
author = {Lopho, Carlos Chavez},
|
31 |
+
title = {TempoFunk: Extending latent diffusion image models to Video},
|
32 |
+
url = {tempofunk.github.io},
|
33 |
+
month = {5},
|
34 |
+
year = {2023}
|
35 |
+
}
|
36 |
+
""",
|
37 |
+
license="AGPL v3"
|
38 |
+
)
|
39 |
+
|
40 |
+
def _split_generators(self, dl_manager):
|
41 |
+
# Load the chunk list.
|
42 |
+
print("PATH:", dl_manager.download("lists/chunk_list.json"))
|
43 |
+
thing = json.load(open(dl_manager.download("lists/chunk_list.json"), 'rb'))
|
44 |
+
_CHUNK_LIST = thing
|
45 |
+
|
46 |
+
# Create a list to hold the downloaded chunks.
|
47 |
+
_list = []
|
48 |
+
|
49 |
+
# Download each chunk file.
|
50 |
+
for chunk in _CHUNK_LIST:
|
51 |
+
_list.append(dl_manager.download(f"data/{chunk}.tar"))
|
52 |
+
|
53 |
+
# Return the list of downloaded chunks.
|
54 |
+
return [
|
55 |
+
datasets.SplitGenerator(
|
56 |
+
name=datasets.Split.TRAIN,
|
57 |
+
gen_kwargs={
|
58 |
+
"chunks": _list,
|
59 |
+
},
|
60 |
+
),
|
61 |
+
]
|
62 |
+
|
63 |
+
def _generate_examples(self, chunks):
|
64 |
+
"""Generate images and labels for splits."""
|
65 |
+
for chunk in chunks:
|
66 |
+
tar_data = open(chunk, 'rb')
|
67 |
+
tar_bytes = tar_data.read()
|
68 |
+
tar_bytes_io = io.BytesIO(tar_bytes)
|
69 |
+
|
70 |
+
response_dict = {}
|
71 |
+
|
72 |
+
with tarfile.open(fileobj=tar_bytes_io, mode='r') as tar:
|
73 |
+
for file_info in tar:
|
74 |
+
if file_info.isfile():
|
75 |
+
file_name = file_info.name
|
76 |
+
#filename format is typ_id.ext
|
77 |
+
file_type = file_name.split('_')[0]
|
78 |
+
file_id = file_name.split('_')[1].split('.')[0]
|
79 |
+
file_ext = file_name.split('_')[1].split('.')[1]
|
80 |
+
file_contents = tar.extractfile(file_info).read()
|
81 |
+
|
82 |
+
if file_id not in response_dict:
|
83 |
+
response_dict[file_id] = {}
|
84 |
+
|
85 |
+
if file_type == 'txt' or file_type == 'vid':
|
86 |
+
response_dict[file_id][file_type] = numpy.load(io.BytesIO(file_contents))
|
87 |
+
elif file_type == 'jso':
|
88 |
+
response_dict[file_id][file_type] = json.loads(file_contents)
|
89 |
+
|
90 |
+
for key, value in response_dict.items():
|
91 |
+
yield key, {
|
92 |
+
"id": key,
|
93 |
+
"description": value['jso']['description'],
|
94 |
+
"prompt": value['txt'],
|
95 |
+
"video": value['vid'],
|
96 |
+
"videourl": value['jso']['videourl'],
|
97 |
+
"categories": value['jso']['categories'],
|
98 |
+
"duration": value['jso']['duration'],
|
99 |
+
"full_metadata": value['jso']
|
100 |
+
}
|