Datasets:

lefreud
/

GimbalDiffusion-PanoWAN-metadata

Exception:    SplitsNotFoundError
Message:      The split names could not be parsed from the dataset config.
Traceback:    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/webdataset/webdataset.py", line 81, in _split_generators
                  first_examples = list(islice(pipeline, self.NUM_EXAMPLES_FOR_FEATURES_INFERENCE))
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/webdataset/webdataset.py", line 32, in _get_pipeline_from_tar
                  fs: fsspec.AbstractFileSystem = fsspec.filesystem("memory")
                                                  ~~~~~~~~~~~~~~~~~^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/fsspec/registry.py", line 302, in filesystem
                  cls = get_filesystem_class(protocol)
                File "/usr/local/lib/python3.14/site-packages/fsspec/registry.py", line 239, in get_filesystem_class
                  raise ValueError(f"Protocol not known: {protocol}")
              ValueError: Protocol not known: memory
              
              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.

GimbalDiffusion PanoWAN Annotations

Filtering results and derived camera/gravity annotations used to train GimbalDiffusion: Gravity-Aware Camera Control for Video Generation.

Source videos are not redistributed. The filenames encode a source video identifier and temporal range; download_clips.py provides best-effort reconstruction for users who are authorized to download the corresponding source material.

Path	Contents
`panowan_filtering_status.csv`	Filtering results for 7,797 clips. Only the 4,798 rows with `status == "valid_sfm"` are retained for training.
`vipe_output.tar`	Archive containing 4,798 retained VIPE camera-pose sequences under `vipe_output/pose/`, stored under the `data` key.
`pfield_estimations.tar`	Archive containing 4,798 retained panorama gravity estimates under `pfield_estimations/`, sampled at eight yaw angles.
`download_clips.py`	Optional source-clip reconstruction helper.

Every retained clip has both a pose and pfield file. Unused annotations, VIPE intrinsics, masks, intermediate outputs, and PanFlow data are deliberately excluded.

Download

hf download lefreud/GimbalDiffusion-PanoWAN-metadata \
  --repo-type dataset \
  --local-dir data/panowan_annotated
tar -xf data/panowan_annotated/vipe_output.tar \
  -C data/panowan_annotated
tar -xf data/panowan_annotated/pfield_estimations.tar \
  -C data/panowan_annotated

The archives unpack the VIPE poses and pfield estimates into the directory layout expected by the dataset reader.

Reconstruct `clips_16_fps`

Install yt-dlp and ffmpeg, then run:

python download_clips.py --root .

The script defaults to the 4,798 filtered training clips, cuts the time range encoded by names such as <video-id>_85s_95s.mp4, converts to 16 fps, writes clips_16_fps/, skips completed files, and continues after unavailable sources. Use --dry-run to inspect the selection or --limit N for a small trial; the limit counts successful clips, so unavailable sources are skipped.

You are responsible for complying with the source platform's terms, copyright, privacy, and any access restrictions. Some videos may have been removed or may not be available in every region. The original PanoWan dataset is the source of the filename/caption metadata.

Training selection

Only rows with status == "valid_sfm" are kept. The GimbalDiffusion dataset reader applies this status selection and captions projected perspective crops through InternVL3 at training time.

License and citation

The original PanoWan dataset is published under Apache-2.0. GimbalDiffusion filtering results and derived pose/gravity annotations in this repository are also released under Apache-2.0; see LICENSE.md. Source videos retain their original terms and are not covered by this repository's license.

@inproceedings{fortierchouinard2026gimbaldiffusion,
  title = {GimbalDiffusion: Gravity-Aware Camera Control for Video Generation},
  author = {Fortier-Chouinard, Fr{\'e}d{\'e}ric and Hold-Geoffroy, Yannick and Deschaintre, Valentin and Gadelha, Matheus and Lalonde, Jean-Fran{\c{c}}ois},
  booktitle = {European Conference on Computer Vision (ECCV)},
  year = {2026}
}