Leyo commited on
Commit
69fab37
1 Parent(s): 406370c

create loading script

Browse files
Files changed (1) hide show
  1. TGIF.py +117 -0
TGIF.py ADDED
@@ -0,0 +1,117 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import csv
2
+ import datasets
3
+ import os
4
+ import urllib.request
5
+
6
+ _CITATION = """
7
+ @InProceedings{tgif-cvpr2016,
8
+ author = {Li, Yuncheng and Song, Yale and Cao, Liangliang and Tetreault, Joel and Goldberg, Larry and Jaimes, Alejandro and Luo, Jiebo},
9
+ title = "{TGIF: A New Dataset and Benchmark on Animated GIF Description}",
10
+ booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
11
+ month = {June},
12
+ year = {2016}
13
+ }
14
+ """
15
+
16
+ _DESCRIPTION = """\
17
+ The Tumblr GIF (TGIF) dataset contains 100K animated GIFs and 120K sentences describing visual content of the animated GIFs.
18
+ The animated GIFs have been collected from Tumblr, from randomly selected posts published between May and June of 2015.
19
+ We provide the URLs of animated GIFs in this release. The sentences are collected via crowdsourcing, with a carefully designed
20
+ annotationinterface that ensures high quality dataset. We provide one sentence per animated GIF for the training and validation splits,
21
+ and three sentences per GIF for the test split. The dataset shall be used to evaluate animated GIF/video description techniques.
22
+ """
23
+
24
+ _URL_BASE = "http://raingo.github.io/TGIF-Release/"
25
+
26
+ _DL_URL = "https://github.com/raingo/TGIF-Release/archive/master.zip"
27
+
28
+
29
+ class TGIFConfig(datasets.BuilderConfig):
30
+ """BuilderConfig for TGIF."""
31
+
32
+ def __init__(self, **kwargs):
33
+ super(TGIFConfig, self).__init__(
34
+ version=datasets.Version("2.1.0", ""), **kwargs)
35
+
36
+
37
+ class TGIF(datasets.GeneratorBasedBuilder):
38
+
39
+ DEFAULT_CONFIG_NAME = "all"
40
+ BUILDER_CONFIGS = [
41
+ TGIFConfig(name="all", description="All the TGIF dataset"),
42
+ ]
43
+
44
+ def _info(self):
45
+ return datasets.DatasetInfo(
46
+ description=_DESCRIPTION,
47
+ features=datasets.Features(
48
+ {
49
+ "video_path": datasets.Value("string"),
50
+ "video_bytes": datasets.Value("large_binary"),
51
+ "en_global_captions": datasets.features.Sequence(datasets.Value("string"))
52
+ }
53
+ ),
54
+ supervised_keys=None,
55
+ homepage=_URL_BASE,
56
+ citation=_CITATION,
57
+ )
58
+
59
+ def _split_generators(self, dl_manager):
60
+ archive_path = dl_manager.download_and_extract(_DL_URL)
61
+ archive_data_path = os.path.join(
62
+ archive_path, "TGIF-Release-master/data/splits/")
63
+ infos_file = os.path.join(
64
+ archive_path, "TGIF-Release-master/data/tgif-v1.0.tsv")
65
+
66
+ train_splits = [
67
+ datasets.SplitGenerator(
68
+ name=datasets.Split.TRAIN,
69
+ gen_kwargs={
70
+ "split_links_file": os.path.join(archive_data_path, "train.txt"),
71
+ "infos_file": infos_file
72
+ },
73
+ )
74
+ ]
75
+ dev_splits = [
76
+ datasets.SplitGenerator(
77
+ name=datasets.Split.VALIDATION,
78
+ gen_kwargs={
79
+ "split_links_file": os.path.join(archive_data_path, "val.txt"),
80
+ "infos_file": infos_file
81
+ },
82
+ )
83
+ ]
84
+ test_splits = [
85
+ datasets.SplitGenerator(
86
+ name=datasets.Split.TEST,
87
+ gen_kwargs={
88
+ "split_links_file": os.path.join(archive_data_path, "test.txt"),
89
+ "infos_file": infos_file
90
+ },
91
+ )
92
+ ]
93
+ return train_splits + dev_splits + test_splits
94
+
95
+ def _generate_examples(self, split_links_file, infos_file):
96
+ """This function returns the examples."""
97
+
98
+ dict = {}
99
+ with open(split_links_file, encoding="utf-8") as txt_file:
100
+ for line in txt_file:
101
+ line = line[0:-1]
102
+ dict[line] = []
103
+ with open(infos_file, encoding="utf-8") as tsv_file:
104
+ tsv_reader = csv.reader(tsv_file, delimiter="\t", quotechar='"')
105
+ for idx, (video_link, text) in enumerate(tsv_reader):
106
+ try:
107
+ dict[video_link].append(text)
108
+ except Exception:
109
+ pass
110
+ for idx, video_link in enumerate(dict):
111
+ video_data = urllib.request.urlopen(video_link).read()
112
+ video_bytes = bytearray(video_data)
113
+ yield idx, {
114
+ "video_path": video_link,
115
+ "video_bytes": video_bytes,
116
+ "en_global_captions": dict[video_link],
117
+ }