Datasets:
fix filename
Browse files- ActivityNet_Captions.py +127 -0
ActivityNet_Captions.py
ADDED
@@ -0,0 +1,127 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Lint as: python3
|
2 |
+
"""TGIF: A New Dataset and Benchmark on Animated GIF Description"""
|
3 |
+
|
4 |
+
import os
|
5 |
+
import json
|
6 |
+
import datasets
|
7 |
+
|
8 |
+
_CITATION = """
|
9 |
+
@inproceedings{krishna2017dense,
|
10 |
+
title={Dense-Captioning Events in Videos},
|
11 |
+
author={Krishna, Ranjay and Hata, Kenji and Ren, Frederic and Fei-Fei, Li and Niebles, Juan Carlos},
|
12 |
+
booktitle={International Conference on Computer Vision (ICCV)},
|
13 |
+
year={2017}
|
14 |
+
}
|
15 |
+
"""
|
16 |
+
|
17 |
+
_DESCRIPTION = """\
|
18 |
+
The ActivityNet Captions dataset connects videos to a series of temporally annotated sentence descriptions.
|
19 |
+
Each sentence covers an unique segment of the video, describing multiple events that occur. These events
|
20 |
+
may occur over very long or short periods of time and are not limited in any capacity, allowing them to
|
21 |
+
co-occur. On average, each of the 20k videos contains 3.65 temporally localized sentences, resulting in
|
22 |
+
a total of 100k sentences. We find that the number of sentences per video follows a relatively normal
|
23 |
+
distribution. Furthermore, as the video duration increases, the number of sentences also increases.
|
24 |
+
Each sentence has an average length of 13.48 words, which is also normally distributed. You can find more
|
25 |
+
details of the dataset under the ActivityNet Captions Dataset section, and under supplementary materials
|
26 |
+
in the paper.
|
27 |
+
"""
|
28 |
+
|
29 |
+
_URL_BASE = "https://cs.stanford.edu/people/ranjaykrishna/densevid/"
|
30 |
+
|
31 |
+
_DL_URL = "https://huggingface.co/datasets/Leyo/ActivityNet_Captions/resolve/main/captions.tar.gz"
|
32 |
+
|
33 |
+
class ActivityNetConfig(datasets.BuilderConfig):
|
34 |
+
"""BuilderConfig for ActivityNet Captions."""
|
35 |
+
|
36 |
+
def __init__(self, **kwargs):
|
37 |
+
super(ActivityNetConfig, self).__init__(
|
38 |
+
version=datasets.Version("2.1.0", ""), **kwargs)
|
39 |
+
|
40 |
+
|
41 |
+
class ActivityNet(datasets.GeneratorBasedBuilder):
|
42 |
+
|
43 |
+
DEFAULT_CONFIG_NAME = "all"
|
44 |
+
BUILDER_CONFIGS = [
|
45 |
+
ActivityNetConfig(
|
46 |
+
name="all", description="All the ActivityNet Captions dataset"),
|
47 |
+
]
|
48 |
+
|
49 |
+
def _info(self):
|
50 |
+
return datasets.DatasetInfo(
|
51 |
+
description=_DESCRIPTION,
|
52 |
+
features=datasets.Features(
|
53 |
+
{
|
54 |
+
"video_id": datasets.Value("string"),
|
55 |
+
"path": datasets.Value("string"),
|
56 |
+
"duration": datasets.Value("float32"),
|
57 |
+
"starts": datasets.features.Sequence(datasets.Value("float32")),
|
58 |
+
"ends": datasets.features.Sequence(datasets.Value("float32")),
|
59 |
+
"captions": datasets.features.Sequence(datasets.Value("string"))
|
60 |
+
}
|
61 |
+
),
|
62 |
+
supervised_keys=None,
|
63 |
+
homepage=_URL_BASE,
|
64 |
+
citation=_CITATION,
|
65 |
+
)
|
66 |
+
|
67 |
+
def _split_generators(self, dl_manager):
|
68 |
+
archive_path = dl_manager.download(_DL_URL)
|
69 |
+
|
70 |
+
train_splits = [
|
71 |
+
datasets.SplitGenerator(
|
72 |
+
name=datasets.Split.TRAIN,
|
73 |
+
gen_kwargs={
|
74 |
+
"files": dl_manager.iterable(archive_path),
|
75 |
+
"ids_file": os.path.join(archive_path, "train_ids.json"),
|
76 |
+
"infos_file": os.path.join(archive_path, "train.json")
|
77 |
+
},
|
78 |
+
)
|
79 |
+
]
|
80 |
+
dev_splits = [
|
81 |
+
datasets.SplitGenerator(
|
82 |
+
name=datasets.Split.VALIDATION,
|
83 |
+
gen_kwargs={
|
84 |
+
"files": dl_manager.iterable(archive_path),
|
85 |
+
"ids_file": os.path.join(archive_path, "val_ids.json"),
|
86 |
+
"infos_file": os.path.join(archive_path, "val_1.json")
|
87 |
+
},
|
88 |
+
)
|
89 |
+
]
|
90 |
+
test_splits = [
|
91 |
+
datasets.SplitGenerator(
|
92 |
+
name=datasets.Split.TEST,
|
93 |
+
gen_kwargs={
|
94 |
+
"files": dl_manager.iterable(archive_path),
|
95 |
+
"ids_file": os.path.join(archive_path, "test_ids.json"),
|
96 |
+
"infos_file": os.path.join(archive_path, "val_2.json")
|
97 |
+
},
|
98 |
+
)
|
99 |
+
]
|
100 |
+
return train_splits + dev_splits + test_splits
|
101 |
+
|
102 |
+
def _generate_examples(self, files, ids_file, infos_file):
|
103 |
+
"""This function returns the examples."""
|
104 |
+
|
105 |
+
for path, f in files:
|
106 |
+
if path == infos_file:
|
107 |
+
with open(f, encoding="utf-8") as json_file:
|
108 |
+
infos = json.load(json_file)
|
109 |
+
for path, f in files:
|
110 |
+
if path == ids_file:
|
111 |
+
with open(f, encoding="utf-8") as json_file:
|
112 |
+
ids = json.load(json_file)
|
113 |
+
for idx, id in enumerate(ids):
|
114 |
+
path = "https://www.youtube.com/watch?v=" + id[2:]
|
115 |
+
starts = [timestamp[0]
|
116 |
+
for timestamp in infos[id]["timestamps"]]
|
117 |
+
ends = [timestamp[1] for timestamp in infos[id]["timestamps"]]
|
118 |
+
yield idx, {
|
119 |
+
"video_id": id,
|
120 |
+
"path": path,
|
121 |
+
"video_id": datasets.Value("string"),
|
122 |
+
"path": datasets.Value("string"),
|
123 |
+
"duration": infos[id]["duration"],
|
124 |
+
"starts": starts,
|
125 |
+
"ends": ends,
|
126 |
+
"captions": infos[id]["sentences"],
|
127 |
+
}
|