The dataset viewer is not available for this split.
Error code: StreamingRowsError
Exception: CastError
Message: Couldn't cast
md5Checksum: string
mimeType: string
size: int64
id: string
name: string
folder: string
relative_path: string
duplicate_ids: list<item: string>
child 0, item: string
AudioFeature: list<item: struct<name: string, selected_id: string, duplicate_id: string, md5Checksum: string>>
child 0, item: struct<name: string, selected_id: string, duplicate_id: string, md5Checksum: string>
child 0, name: string
child 1, selected_id: string
child 2, duplicate_id: string
child 3, md5Checksum: string
VideoFeature: list<item: null>
child 0, item: null
to
{'AudioFeature': List({'name': Value('string'), 'selected_id': Value('string'), 'duplicate_id': Value('string'), 'md5Checksum': Value('string')}), 'VideoFeature': List(Value('null'))}
because column names don't match
Traceback: Traceback (most recent call last):
File "/src/services/worker/src/worker/utils.py", line 147, in get_rows_or_raise
return get_rows(
dataset=dataset,
...<4 lines>...
column_names=column_names,
)
File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
return func(*args, **kwargs)
File "/src/services/worker/src/worker/utils.py", line 127, in get_rows
rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
File "/src/services/worker/src/worker/utils.py", line 483, in safe_iter
yield from ds.decode(False) if ds.features else ds
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2951, in __iter__
for key, example in ex_iterable:
^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2461, in __iter__
for key, pa_table in self._iter_arrow():
~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2486, in _iter_arrow
for key, pa_table in self.ex_iterable._iter_arrow():
~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 547, in _iter_arrow
for key, pa_table in iterator:
^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 430, in _iter_arrow
for key, pa_table in self.generate_tables_fn(**gen_kwags):
~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 343, in _generate_tables
self._cast_table(pa_table, json_field_paths=json_field_paths),
~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
pa_table = table_cast(pa_table, self.info.features.arrow_schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2378, in table_cast
return cast_table_to_schema(table, schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2306, in cast_table_to_schema
raise CastError(
...<3 lines>...
)
datasets.table.CastError: Couldn't cast
md5Checksum: string
mimeType: string
size: int64
id: string
name: string
folder: string
relative_path: string
duplicate_ids: list<item: string>
child 0, item: string
AudioFeature: list<item: struct<name: string, selected_id: string, duplicate_id: string, md5Checksum: string>>
child 0, item: struct<name: string, selected_id: string, duplicate_id: string, md5Checksum: string>
child 0, name: string
child 1, selected_id: string
child 2, duplicate_id: string
child 3, md5Checksum: string
VideoFeature: list<item: null>
child 0, item: null
to
{'AudioFeature': List({'name': Value('string'), 'selected_id': Value('string'), 'duplicate_id': Value('string'), 'md5Checksum': Value('string')}), 'VideoFeature': List(Value('null'))}
because column names don't matchNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
MSRVTT ImageBind teacher features
This dataset mirrors the precomputed MSRVTT audio and video teacher features from the source Google Drive folders below. It contains original PyTorch .pt files, without re-encoding or tensor conversion.
| Directory | Unique files | Video IDs | Uncompressed bytes |
|---|---|---|---|
AudioFeature |
10,000 | video0.pt to video9999.pt |
48,486,030 |
VideoFeature |
10,000 | video0.pt to video9999.pt |
48,486,030 |
| Total | 20,000 | 10,000 paired videos | 96,972,060 |
Sources
The source folders were inventoried on 2026-10-17. The audio folder contains 10,022 Drive entries with 10,000 unique filenames: 22 filenames have a second entry with identical size and MD5. One copy of each identical file is retained, and the additional Drive IDs are recorded in source-manifest.json and source-duplicates.json. No conflicting versions were found. The video folder contains 10,000 entries.
Layout
AudioFeature/video0.pt ... video9999.pt
VideoFeature/video0.pt ... video9999.pt
source-manifest.json
source-duplicates.json
SHA256SUMS.txt
README.md
Every downloaded feature is checked against the file size and MD5 supplied by Google Drive. SHA256SUMS.txt records SHA-256 hashes of the mirrored feature files. Filenames and bytes are preserved.
Download and use
from huggingface_hub import snapshot_download
feature_dir = snapshot_download(
repo_id="RichyHu999/msrvtt-imagebind-features",
repo_type="dataset",
local_dir="data/msrvtt/FeatureData/ImageBind",
)
This repository is public; signing in is not required for downloads.
For the full supplied collection, including MSRVTT annotations and ASR plus Charades, VATEX and LSMDC source files, see Multimodal Teacher Data.
from pathlib import Path
import torch
root = Path(feature_dir)
audio = torch.load(root / "AudioFeature/video0.pt", map_location="cpu", weights_only=True)
video = torch.load(root / "VideoFeature/video0.pt", map_location="cpu", weights_only=True)
The mirror contains features only. It does not contain raw videos, audio recordings, captions, split definitions, or features from Charades, LSMDC, or VATEX. It does not independently establish the upstream feature extraction settings. No new license is granted by this mirror; refer to the original dataset and feature providers for applicable terms and attribution.
- Downloads last month
- 28