Commit
·
d0a86b3
1
Parent(s):
48d87c6
update: builder
Browse files- how2sign-clips.py +32 -11
how2sign-clips.py
CHANGED
|
@@ -15,7 +15,7 @@ _DESCRIPTION = """
|
|
| 15 |
_HOMEPAGE = "https://how2sign.github.io/index.html"
|
| 16 |
_REPO_URL = "https://huggingface.co/datasets/VieSignLang/how2sign-clips/resolve/main"
|
| 17 |
_URLS = {
|
| 18 |
-
"meta": f"{_REPO_URL}/metadata/" + "
|
| 19 |
"video": f"{_REPO_URL}/videos/" + "{type}/{split}/*/*.zip",
|
| 20 |
}
|
| 21 |
|
|
@@ -46,12 +46,22 @@ class How2SignClips(datasets.GeneratorBasedBuilder):
|
|
| 46 |
|
| 47 |
def _info(self) -> datasets.DatasetInfo:
|
| 48 |
features = datasets.Features({
|
| 49 |
-
"
|
| 50 |
-
"
|
| 51 |
-
"
|
| 52 |
-
"
|
|
|
|
|
|
|
|
|
|
| 53 |
"video": datasets.Value("large_binary"),
|
| 54 |
})
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 55 |
|
| 56 |
return datasets.DatasetInfo(
|
| 57 |
description=_DESCRIPTION,
|
|
@@ -127,16 +137,27 @@ class How2SignClips(datasets.GeneratorBasedBuilder):
|
|
| 127 |
video_path = os.path.join(
|
| 128 |
video_dir,
|
| 129 |
f"shard_{shard_idx:03d}_{num_shards:03d}",
|
| 130 |
-
sample["
|
|
|
|
| 131 |
)
|
| 132 |
if os.path.exists(video_path):
|
| 133 |
yield i, {
|
| 134 |
-
"
|
| 135 |
-
"
|
| 136 |
-
"
|
| 137 |
-
"
|
| 138 |
-
"
|
|
|
|
|
|
|
|
|
|
| 139 |
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 140 |
|
| 141 |
def __get_binary_data(self, path):
|
| 142 |
with open(path, "rb") as f:
|
|
|
|
| 15 |
_HOMEPAGE = "https://how2sign.github.io/index.html"
|
| 16 |
_REPO_URL = "https://huggingface.co/datasets/VieSignLang/how2sign-clips/resolve/main"
|
| 17 |
_URLS = {
|
| 18 |
+
"meta": f"{_REPO_URL}/metadata/" + "sharded_how2sign_realigned_{split}.parquet",
|
| 19 |
"video": f"{_REPO_URL}/videos/" + "{type}/{split}/*/*.zip",
|
| 20 |
}
|
| 21 |
|
|
|
|
| 46 |
|
| 47 |
def _info(self) -> datasets.DatasetInfo:
|
| 48 |
features = datasets.Features({
|
| 49 |
+
"VIDEO_ID": datasets.Value("string"),
|
| 50 |
+
"VIDEO_NAME": datasets.Value("string"),
|
| 51 |
+
"SENTENCE_ID": datasets.Value("string"),
|
| 52 |
+
"SENTENCE_NAME": datasets.Value("string"),
|
| 53 |
+
"START_REALIGNED": datasets.Value("float64"),
|
| 54 |
+
"END_REALIGNED": datasets.Value("float64"),
|
| 55 |
+
"SENTENCE": datasets.Value("string"),
|
| 56 |
"video": datasets.Value("large_binary"),
|
| 57 |
})
|
| 58 |
+
# features = datasets.Features({
|
| 59 |
+
# "id": datasets.Value("string"),
|
| 60 |
+
# "type": datasets.Value("string"),
|
| 61 |
+
# "view": datasets.Value("string"),
|
| 62 |
+
# "text": datasets.Value("string"),
|
| 63 |
+
# "video": datasets.Value("large_binary"),
|
| 64 |
+
# })
|
| 65 |
|
| 66 |
return datasets.DatasetInfo(
|
| 67 |
description=_DESCRIPTION,
|
|
|
|
| 137 |
video_path = os.path.join(
|
| 138 |
video_dir,
|
| 139 |
f"shard_{shard_idx:03d}_{num_shards:03d}",
|
| 140 |
+
sample["SENTENCE_NAME"] + ".mp4",
|
| 141 |
+
# sample["id"] + ".mp4",
|
| 142 |
)
|
| 143 |
if os.path.exists(video_path):
|
| 144 |
yield i, {
|
| 145 |
+
"VIDEO_ID": sample["VIDEO_ID"],
|
| 146 |
+
"VIDEO_NAME": sample["VIDEO_NAME"],
|
| 147 |
+
"SENTENCE_ID": sample["SENTENCE_ID"],
|
| 148 |
+
"SENTENCE_NAME": sample["SENTENCE_NAME"],
|
| 149 |
+
"START_REALIGNED": sample["START_REALIGNED"],
|
| 150 |
+
"END_REALIGNED": sample["END_REALIGNED"],
|
| 151 |
+
"SENTENCE": sample["SENTENCE"],
|
| 152 |
+
"VIDEO": self.__get_binary_data(video_path),
|
| 153 |
}
|
| 154 |
+
# yield i, {
|
| 155 |
+
# "id": sample["id"],
|
| 156 |
+
# "type": sample["type"],
|
| 157 |
+
# "view": sample["view"],
|
| 158 |
+
# "text": sample["text"],
|
| 159 |
+
# "video": self.__get_binary_data(video_path),
|
| 160 |
+
# }
|
| 161 |
|
| 162 |
def __get_binary_data(self, path):
|
| 163 |
with open(path, "rb") as f:
|