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 "/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 84, in _split_generators
                  raise ValueError(
                  ...<2 lines>...
                  )
              ValueError: The TAR archives of the dataset should be in WebDataset format, but the files in the archive don't share the same prefix or the same types.
              
              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 68, 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.

jiggle2-layers-wm

World-model training and evaluation samples derived from x5liang/jiggle2-layers revision v2 (layered-tissue scene: a flap is lifted, marked and closed, and the marks must be recalled when it is lifted again). These are exactly the samples our memory-augmented world models (Wan2.2-TI2V-5B backbone) were trained and evaluated on, so that runs elsewhere use identical inputs.

Made with git tag v0.3.0 of the jiggle2 code (surgwm.layers precompute / frames / keyframes, surgwm.warp_memory, surgwm.source_map), packed with python -m surgwm.memory.hpmc.warp_attn_study pack.

Layout

One uncompressed tar per episode in episodes/<ep>.tar; paths inside are relative to the samples root:

path content
clips/<ep>_t<t0>.pt one 81-frame clip from frame t0: VAE latents (48 × 21 × 16 × 20), per-frame robot actions and camera, tool silhouettes, meta (16 clips per episode: probe, touch, uniform)
clips/<ep>.done completion marker
frames/<ep>.pt single-frame VAE latents of every 4th frame (memory frames)
keyframes/<ep>.npz per-frame statistics used to choose memory frames (mark visibility, surface coverage, zoom)
warp/<ep>/{rgb,depth,tool,K,c2w}.npy every 4th frame at 320 × 240 with depth and tool mask, intrinsics, camera poses (warping memory into future views)
source/<ep>.npz per-vertex source tables for mark scoring and source-map labels (584 episodes)

Plus prompt_embeds.pt (umT5 embedding of the fixed text prompt) and manifest.json (files per episode).

624 episodes, 84 GB. 584 are built from jiggle2-layers v2 as published. 40 come from the earlier copies of the episodes whose camera jumped, which were regenerated before v2 was published (their raw episodes are not on the Hub and they have no source tables). The train / held-out split is configs/wm_layers_v2_split.json in the code.

Use

python -m surgwm.memory.hpmc.warp_attn_study fetch --samples /data/samples      # download, unpack and check, episode by episode

License: CC BY 4.0, as the source dataset.

Downloads last month
278