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.

Diogenes Robotics Warehouse Iterations Archive v02

Historical synthetic capture experiments from the Apache-2.0 Robotics-Warehouse scene: 184 MP4 videos, 46 WAV recordings, 1,114 JSON metadata files, 84 PNG inspection sheets, 150 logs/XML/Markdown reports and 262 status-marker files. Seven iteration collections contain 1,840 archived objects / 17,481,781,961 bytes after privacy edits. Versions overlap; these counts do not imply independent scenes, a quality-filtered dataset or a train/test split.

Collection MP4 files All objects Archived bytes
warehouse-try1 4 41 381,170,646
warehouse-v1 44 459 4,194,136,373
warehouse-v2 40 399 3,832,241,318
warehouse-v3 44 441 4,184,428,559
warehouse-v4 48 455 4,580,893,504
warehouse-v5 4 36 308,910,147
warehouse-v6 0 9 1,414

Layout and use

Tar members preserve paths under collections/<iteration>/. Completed clip directories generally include four video lanes, action-camera arrays, system information, capture provenance, the input plan and available co-visibility/dynamic-object metrics. PNGs in visual-audit are original inspection contact sheets. Failed planning attempts may contain only a status marker; warehouse-v6 contains nine such failures, not missing video payloads.

Action arrays include frame/time/fps, camera position, orientation, intrinsics and speed, with version-dependent generated player/input fields. player_* can mirror camera pose or its ground projection; these are not real human controls. Euler ordering and other convention caveats are retained. Read each file's field_conventions before comparing runs. Nominal clips are 60 seconds at 30 fps; collection plans include 16:9, 9:16, 4:3 and 3:4 shapes. All sampled videos were 1920×1080; consult individual files and historical aspect failures instead of assuming the plans were realized exactly.

Video, WAV, PNG and numeric trajectory bytes are preserved. Existing audio is engine-rendered or explicitly authored procedural ambience. The later source scene had no native AudioSource: its provenance identifies one room-tone and three machinery sources authored by Diogenes. It is not a natural-world acoustic recording.

import json, tarfile
with tarfile.open("downloaded-shard.tar", "r:") as archive:
    for member in archive:
        if member.isfile() and member.name.endswith("qc-report.json"):
            record = json.load(archive.extractfile(member))
            print(member.name, record.get("verdict"), record.get("counts"))

Historical defects remain defects

Early iterations warehouse-v1 and warehouse-v2 incorrectly label scene and process identity as FactoryTraining. warehouse-v3 identifies the warehouse but retains the wrong Asset Store licence marker. The project documented that their Unity logs actually loaded Assets/Warehouse/Scenes/Warehouse.unity; the underlying assets were compared with the pinned Apache-2.0 upstream. Original wrong labels and all 68 VOID markers are deliberately preserved. Do not use those labels as authoritative scene/licence metadata or treat the archive as reviving the old delivery approvals. warehouse-v4 introduced corrected provenance; warehouse-v5 also includes a visibly failed run with the warehouse environment absent. Blank visual-audit checklists, QA failures, numerical PASS and failed planning are all retained. No new production certification or training eligibility is claimed.

These are useful records of camera planning, synthetic data failures, provenance drift and QA limits. See the documented correction in project source. The old operational instruction against uploading VOID batches remains historical text; the project owner separately authorized this explicitly labelled complete research archive.

Privacy and verification

All 1,114 JSON files were fully parsed with literal field/type and string review; all 150 technical text files and 262 marker contents were read. Representative visual review covered 40 video first frames and 10 original inspection sheets across all video-bearing iterations. No private account or desktop overlay was identified in those samples; this is not exhaustive video viewing or audio transcription.

234 explicit public copies replace 96 JSON files containing local paths and 138 log/XML files containing local paths or Unity licensing/machine/session diagnostics. Personal paths become archive-relative references or labelled placeholders; specified licensing diagnostic lines and runtime fields are replaced with labelled placeholders. Version v02 additionally replaces the identified Unity machine/network diagnostic lines. JSON numeric/boolean values and array structures are unchanged, XML structure and technical test results are retained, and technical text line counts are preserved. Original files are privately retained and SHA-256 verified. PUBLIC-COPIES.json distinguishes historical original hashes from the actual public hashes; the archive manifest must be used to verify public bytes. Unmodified files remain byte-for-byte original.

Publication is complete: all 1,840 public objects are preserved in 20 TAR shards. MANIFEST.jsonl gives every member's actual size, SHA-256 and containing archive; ARCHIVE_INDEX.jsonl gives each shard's immutable upload revision, size and SHA-256. All shards were checked against pinned Hugging Face LFS and anonymous range downloads. The rebuilt metadata shard also received anonymous full-byte SHA-256 verification. The rebuilt metadata shard also received anonymous full-byte SHA-256 verification. All original capture objects and temporary public copies have been confirmed absent from R2; the 234 private originals and their transformation evidence remain in the private audit.

Attribution

Robotics-Warehouse, Copyright 2021 Unity Technologies, licensed under the Apache License, Version 2.0. The used base/URP assets came from commit 4a2a506dd48c62bbd1ee5de4eb15ddd679a38e6f; the project recorded 88 unchanged asset files and 22 material serialization upgrades by Unity 6000.5.6f1. See NOTICE.md and the accompanying Apache licence. No Unity editor executable or Asset Store source package is included.

Corrected publication history

This v02 repository has a clean publication history. It preserves all scientific payloads from the withdrawn v01, with 46 Unity logs receiving an additional explicit machine/network diagnostic-line correction. PRIVACY-CORRECTIONS.json preserves the historical-original → v01-public → v02-public SHA-256 chain for every existing publication copy. Line counts and all non-diagnostic bytes are unchanged; this is a defined text correction, not a claim of exhaustive privacy detection. Full old metadata TARs, original private files, and historical documentation remain in the private provenance archive.

Downloads last month
69