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 80, 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.

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

Panorama → Perspective Dataset Pipeline

Code bundle (no video/image data) for the pipeline that crops equirectangular (ERP) panorama source video into multiple perspective/pinhole trajectories, captions the crops, and packages/uploads the result as a streaming HuggingFace dataset. Pulled together from a larger internal codebase for reference.

What's here

qwen36_test/caption_qwen36_8k_5traj.py — the source of the ERP↔pinhole crop math (crop_perspective_with_pitch and friends) used everywhere else in this bundle. For one 8K ERP panorama clip, generates 5 independent trajectories (front_static, out_back_right_first, out_back_left_first, yaw_right, yaw_left), crops each to a perspective video, and captions it with Qwen3.6 via vLLM.
benchmark/make_perspective_from_panorama.py — reuses only the crop math above (no captioning, no GPU) to build a small benchmark GT set: 2 trajectories per clip (out_back_right_first / out_back_left_first), written out as <trajectory>.gt.mp4 + a perspective_meta.json recording the exact per-frame yaw/pitch/fov trajectory used, so the same crop can be reproduced later (e.g. against a model's own generated panorama).
data/dataset_streaming_template/ — the large-scale training-dataset pipeline: streams source panorama shards down from HuggingFace, crops + captions each clip (5-trajectory, and variant scripts for other view counts/configs), uploads the result, deletes the local copy, and can resume a partial run. stream_process_qwen36_8k_5traj.py is the main entry point using the 5-trajectory scheme above; upload_360x_train_dataset.py is a simpler one-shot uploader for an already-processed local dataset. See its own README.md for the dataset shape and shard layout.
data/DATA_LOCATION.md — map of which HuggingFace repos hold which stage of the data (source ERP → warped → cropped+captioned), and which account's token is needed to write to each (tokens shown masked only).

Pipeline flow

enrue1893/PanRealCal (calibrated ERP source)
  -> panorama_warp_processing (warp)
  -> PinPanCalWarp (warped/gt/mask/w2cs, split across enrue1893 + kaistcvlab, ~5214 shards)
  -> stream_process_qwen36_8k_5traj.py (streams shards, 5-trajectory crop + caption)
  -> kaistcvlab/PinPanCalWarp-5traj-captions

benchmark/make_perspective_from_panorama.py branches off the same crop math to build a much smaller, fixed benchmark GT set (2 trajectories/clip) rather than the full training dataset.

Notes / gotchas (from the code's own comments)

Synth-split clips need a yaw sign flip + a fixed -90° offset relative to real clips when reprojecting a model's own generated panorama against this data (CARLA's left-handed heading convention vs. the ERP's own frame) — see panorama_to_perspective_crop.py elsewhere in the parent codebase, which reuses this same crop math for that purpose.
Read access to the public dataset repos needs no token; only write/upload does — see data/DATA_LOCATION.md.

Downloads last month: 33

Total file size:

99.9 kB