Commit
·
e0ba264
1
Parent(s):
56201f1
fix: builder
Browse files- how2sign-clips.py +27 -15
how2sign-clips.py
CHANGED
|
@@ -69,14 +69,23 @@ class How2SignClips(datasets.GeneratorBasedBuilder):
|
|
| 69 |
:return: Splits.
|
| 70 |
"""
|
| 71 |
split_dict = {
|
| 72 |
-
"train":
|
| 73 |
-
|
| 74 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 75 |
}
|
| 76 |
|
| 77 |
return [
|
| 78 |
datasets.SplitGenerator(
|
| 79 |
-
name=name,
|
| 80 |
gen_kwargs={
|
| 81 |
"metadata_path": dl_manager.download(
|
| 82 |
_URLS["meta"].format(split=split)
|
|
@@ -89,28 +98,31 @@ class How2SignClips(datasets.GeneratorBasedBuilder):
|
|
| 89 |
)
|
| 90 |
)
|
| 91 |
),
|
|
|
|
| 92 |
},
|
| 93 |
)
|
| 94 |
-
for split,
|
| 95 |
]
|
| 96 |
|
| 97 |
def _generate_examples(
|
| 98 |
self, metadata_path: str,
|
| 99 |
video_dirs: list[str],
|
|
|
|
| 100 |
) -> tuple[int, dict]:
|
| 101 |
"""
|
| 102 |
Generate examples from metadata.
|
| 103 |
:param metadata_path: Path to metadata.
|
| 104 |
-
:param
|
|
|
|
| 105 |
:yield: Example.
|
| 106 |
"""
|
|
|
|
| 107 |
dataset = datasets.load_dataset(
|
| 108 |
"parquet",
|
| 109 |
data_files=metadata_path,
|
| 110 |
split="train",
|
| 111 |
)
|
| 112 |
for i, sample in enumerate(dataset):
|
| 113 |
-
num_shards = len(video_dirs)
|
| 114 |
shard_idx = sample["shard"]
|
| 115 |
for video_dir in video_dirs:
|
| 116 |
video_path = os.path.join(
|
|
@@ -118,14 +130,14 @@ class How2SignClips(datasets.GeneratorBasedBuilder):
|
|
| 118 |
f"shard_{shard_idx:03d}_{num_shards:03d}",
|
| 119 |
sample["id"] + ".mp4"
|
| 120 |
)
|
| 121 |
-
|
| 122 |
-
|
| 123 |
-
|
| 124 |
-
|
| 125 |
-
|
| 126 |
-
|
| 127 |
-
|
| 128 |
-
|
| 129 |
|
| 130 |
def __get_binary_data(self, path):
|
| 131 |
with open(path, "rb") as f:
|
|
|
|
| 69 |
:return: Splits.
|
| 70 |
"""
|
| 71 |
split_dict = {
|
| 72 |
+
"train": {
|
| 73 |
+
"name": datasets.Split.TRAIN,
|
| 74 |
+
"num_shards": 32,
|
| 75 |
+
},
|
| 76 |
+
"test": {
|
| 77 |
+
"name": datasets.Split.TEST,
|
| 78 |
+
"num_shards": 3,
|
| 79 |
+
},
|
| 80 |
+
"val": {
|
| 81 |
+
"name": datasets.Split.VALIDATION,
|
| 82 |
+
"num_shards": 2,
|
| 83 |
+
},
|
| 84 |
}
|
| 85 |
|
| 86 |
return [
|
| 87 |
datasets.SplitGenerator(
|
| 88 |
+
name=info["name"],
|
| 89 |
gen_kwargs={
|
| 90 |
"metadata_path": dl_manager.download(
|
| 91 |
_URLS["meta"].format(split=split)
|
|
|
|
| 98 |
)
|
| 99 |
)
|
| 100 |
),
|
| 101 |
+
"num_shards": info["num_shards"],
|
| 102 |
},
|
| 103 |
)
|
| 104 |
+
for split, info in split_dict.items()
|
| 105 |
]
|
| 106 |
|
| 107 |
def _generate_examples(
|
| 108 |
self, metadata_path: str,
|
| 109 |
video_dirs: list[str],
|
| 110 |
+
num_shards: int,
|
| 111 |
) -> tuple[int, dict]:
|
| 112 |
"""
|
| 113 |
Generate examples from metadata.
|
| 114 |
:param metadata_path: Path to metadata.
|
| 115 |
+
:param visual_dirs: Directories of videos.
|
| 116 |
+
:param num_shards: Number of shards.
|
| 117 |
:yield: Example.
|
| 118 |
"""
|
| 119 |
+
|
| 120 |
dataset = datasets.load_dataset(
|
| 121 |
"parquet",
|
| 122 |
data_files=metadata_path,
|
| 123 |
split="train",
|
| 124 |
)
|
| 125 |
for i, sample in enumerate(dataset):
|
|
|
|
| 126 |
shard_idx = sample["shard"]
|
| 127 |
for video_dir in video_dirs:
|
| 128 |
video_path = os.path.join(
|
|
|
|
| 130 |
f"shard_{shard_idx:03d}_{num_shards:03d}",
|
| 131 |
sample["id"] + ".mp4"
|
| 132 |
)
|
| 133 |
+
if os.path.exists(video_path):
|
| 134 |
+
yield i, {
|
| 135 |
+
"id": sample["id"],
|
| 136 |
+
"type": sample["type"],
|
| 137 |
+
"view": sample["view"],
|
| 138 |
+
"text": sample["text"],
|
| 139 |
+
"video": self.__get_binary_data(video_path),
|
| 140 |
+
}
|
| 141 |
|
| 142 |
def __get_binary_data(self, path):
|
| 143 |
with open(path, "rb") as f:
|