Dataset Viewer
The dataset viewer is not available for this subset.
Cannot get the split names for the config 'default' of the dataset.
Exception:    SplitsNotFoundError
Message:      The split names could not be parsed from the dataset config.
Traceback:    Traceback (most recent call last):
                File "tsfile/tsfile_py_cpp.pyx", line 567, in tsfile.tsfile_py_cpp.tsfile_reader_new_c
              tsfile.exceptions.FileOpenError: 28: 
              
              The above exception was the direct cause of the following exception:
              
              Traceback (most recent call last):
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 286, in get_dataset_config_info
                  for split_generator in builder._split_generators(
                                         ~~~~~~~~~~~~~~~~~~~~~~~~~^
                      StreamingDownloadManager(base_path=builder.base_path, download_config=download_config)
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                  )
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/tsfile/tsfile.py", line 271, in _split_generators
                  scan = self._scan_metadata(all_files)
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/tsfile/tsfile.py", line 318, in _scan_metadata
                  with self._open_reader(file) as reader:
                       ~~~~~~~~~~~~~~~~~^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/tsfile/tsfile.py", line 742, in _open_reader
                  return TsFileReader(file)
                File "tsfile/tsfile_reader.pyx", line 323, in tsfile.tsfile_reader.TsFileReaderPy.__init__
              SystemError: <class '_weakrefset.WeakSet'> returned a result with an exception set
              
              The above exception was the direct cause of the following exception:
              
              Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/config/split_names.py", line 71, in compute_split_names_from_streaming_response
                  for split in get_dataset_split_names(
                               ~~~~~~~~~~~~~~~~~~~~~~~^
                      path=dataset,
                      ^^^^^^^^^^^^^
                      config_name=config,
                      ^^^^^^^^^^^^^^^^^^^
                      token=hf_token,
                      ^^^^^^^^^^^^^^^
                  )
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 340, in get_dataset_split_names
                  info = get_dataset_config_info(
                      path,
                  ...<6 lines>...
                      **config_kwargs,
                  )
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 291, in get_dataset_config_info
                  raise SplitsNotFoundError("The split names could not be parsed from the dataset config.") from err
              datasets.inspect.SplitsNotFoundError: The split names could not be parsed from the dataset config.

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

Fabric Winding 4-Camera TsFile

Apache TsFile conversion of axiboai/fabric_winding_4cam, a LeRobot v2.1 humanoid upper-body dataset for the task “wind the motors”.

Source and attribution

  • Publishing organization/uploader: AXIBO (axiboai)
  • Individual author: not named in the source card or metadata
  • License: Apache-2.0
  • Paper/citation: not supplied by the source card
  • Split: train; 115 episodes, 188,387 frames, 1 task, 30 fps
  • Robot type: humanoid_upper_body; LeRobot codebase v2.1
  • Task index 0: “wind the motors”

Source Parquet and videos

The source has 115 Parquet files at data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet.

The four streams cam_front, cam_top, cam_left_wrist, and cam_right_wrist contain 115 MP4 files each at videos/chunk-{episode_chunk:03d}/{video_key}/episode_{episode_index:06d}.mp4. Videos are not duplicated in this repository; use the original videos directory. Numeric rows remain aligned through episode_index and frame_index.

Converted artifacts

Train shard Episodes Rows Bytes
fabric_winding_4cam_train-00000-of-00016.tsfile 0-3 (4) 10,452 611,105
fabric_winding_4cam_train-00001-of-00016.tsfile 4-7 (4) 10,239 598,373
fabric_winding_4cam_train-00002-of-00016.tsfile 8-12 (5) 10,364 596,870
fabric_winding_4cam_train-00003-of-00016.tsfile 13-19 (7) 11,738 697,783
fabric_winding_4cam_train-00004-of-00016.tsfile 20-26 (7) 11,224 664,163
fabric_winding_4cam_train-00005-of-00016.tsfile 27-32 (6) 11,102 661,674
fabric_winding_4cam_train-00006-of-00016.tsfile 33-39 (7) 10,957 644,272
fabric_winding_4cam_train-00007-of-00016.tsfile 40-47 (8) 11,745 692,285
fabric_winding_4cam_train-00008-of-00016.tsfile 48-55 (8) 10,970 650,735
fabric_winding_4cam_train-00009-of-00016.tsfile 56-63 (8) 11,418 673,418
fabric_winding_4cam_train-00010-of-00016.tsfile 64-71 (8) 11,551 702,371
fabric_winding_4cam_train-00011-of-00016.tsfile 72-78 (7) 11,659 682,245
fabric_winding_4cam_train-00012-of-00016.tsfile 79-85 (7) 11,004 667,851
fabric_winding_4cam_train-00013-of-00016.tsfile 86-92 (7) 11,354 689,056
fabric_winding_4cam_train-00014-of-00016.tsfile 93-100 (8) 11,565 705,864
fabric_winding_4cam_train-00015-of-00016.tsfile 101-114 (14) 21,045 1,275,193
Total 115 188,387 11,213,258

All shards use the table fabric_winding_4cam_train. The source Parquet files total 17,851,827 bytes; the merged staging Parquet is 12,015,759 bytes.

TsFile schema

Time = round(timestamp * 1000) milliseconds and restarts at zero for each episode. The source timestamp in seconds is dropped because it is redundant with Time / 1000.

TAG columns use the TsFile table device/tag mechanism: episode_index, task_index.

FIELD columns are frame_index, sample_index (renamed from source index), and 36 flattened FLOAT fields:

Index Source dimension State FIELD Action FIELD
0 left_shoulder_pitch observation_state_0 action_0
1 left_shoulder_roll observation_state_1 action_1
2 left_shoulder_yaw observation_state_2 action_2
3 left_elbow observation_state_3 action_3
4 left_wrist_roll observation_state_4 action_4
5 left_wrist_pitch observation_state_5 action_5
6 left_wrist_yaw observation_state_6 action_6
7 right_shoulder_pitch observation_state_7 action_7
8 right_shoulder_roll observation_state_8 action_8
9 right_shoulder_yaw observation_state_9 action_9
10 right_elbow observation_state_10 action_10
11 right_wrist_roll observation_state_11 action_11
12 right_wrist_pitch observation_state_12 action_12
13 right_wrist_yaw observation_state_13 action_13
14 left_hand_lift observation_state_14 action_14
15 left_hand_push observation_state_15 action_15
16 right_hand_lift observation_state_16 action_16
17 right_hand_push observation_state_17 action_17

Physical codec policy:

  • Time: TS_2DIFF + LZ4
  • INT32/INT64: TS_2DIFF + LZ4
  • FLOAT/DOUBLE: GORILLA + LZ4
  • BOOLEAN: RLE + LZ4
  • TAG: TsFile table device/tag storage

Only redundant timestamp is dropped. Camera columns are video references in the source metadata and remain available in the original repository.

Validation

All shards passed exact Python SDK readback against their staging Parquet data and Java schema/codec inspection. Together they contain 188,387 rows and 115 TAG devices with no duplicate TAG/Time rows. Local JSON and Markdown validation reports and the conversion script are intentionally excluded from this repository.

Usage

from tsfile import TsFileReader

reader = TsFileReader("data/fabric_winding_4cam_train-00000-of-00016.tsfile")
columns = [
    "episode_index", "task_index", "frame_index", "sample_index",
    "observation_state_0", "action_0",
]
with reader.query_table("fabric_winding_4cam_train", columns, batch_size=65536) as result:
    print(result.read_arrow_batch().to_pandas().head())
Downloads last month
-