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 Fallout 4 Depth Validation Archive v01

Historical project-captured gameplay, depth-conversion experiments, timing/alignment checks and their technical records. This is a reproducibility archive of experiments, including failed and superseded results. Original non_trainable, local_debug_non_trainable, NOT deliverable and reconstruction labels remain in force; publication is not a new training or delivery certification.

The reviewed public queue contains 9,204 files / 15,129,325,183 bytes, including 64 explicit metadata privacy copies. PATHS.jsonl is the complete logical file inventory.

Contents and interpretation

The collections cover the issue-709 Sanctuary/Vault experiments, a delivery iteration, loop/displacement reconstruction, batch9/compliance checks and seven named validation sessions, including Diamond City and surrounding campaign scenes. They include the original segmented MP4/FLAC or compressed loopback PCM where present, frame/input logs, device/recording profile definitions, raw EXR depth, packaged depth ZIPs, camera/player telemetry, converted 15/30 fps action records, numerical QA, report images, point clouds/meshes and self-contained visualization HTML.

Raw depth-lab records and converted delivery depth are different representations. Raw profiles describe a depth-buffer channel and transformation such as one_minus_d; converted EXRs and camera records use the conventions stated in each run’s metadata. Do not apply one global orientation, unit scale or pose convention to every directory. RGB-D odometry labelled reconstruction is not equivalent to instrumented camera telemetry. Original failures, absent-depth reasons, incorrect/rejected alignments and pending requirements are retained.

Three zero-byte MP4 files inside historical .tmp recording segments are preserved as incomplete capture artefacts. They are not decodable videos and are not counted as successful examples. Input-device handles are salted session-local identifiers from the capture implementation, not published physical device serial numbers. Audio is recorded game-process loopback according to the source metadata; this is not a microphone dataset.

A nested depth.zip from the original FO4 Sanctuary delivery container differs from the loose delivery depth.zip: all 900 same-named EXR payloads differ. The exact bundled version is separately retained at container-variants/FO4Sanctuary-original-bundled/depth.zip. Both variants are preserved, with per-member hashes and provenance; this closure archive does not infer or apply a geometry/orientation repair.

Paths, containers and reuse

PATHS.jsonl maps each stored TAR member to its original capture-relative path. Collection aliases and shortened UUID directory components keep TAR paths portable; the mapping preserves the exact logical original names. Download a shard listed in ARCHIVE_INDEX.jsonl at its pinned revision, then use MANIFEST.jsonl to verify the actual size and SHA-256 of each regular member. Restore original logical folders under a chosen collections/ root when following relative references in historical metadata. Historical R2 URLs describe provenance and may no longer resolve.

import json, tarfile
mapping = {r['path']: r for r in map(json.loads, open('PATHS.jsonl'))}
with tarfile.open('downloaded-shard.tar', 'r:') as archive:
    for member in archive:
        if member.isfile() and member.name in mapping:
            print(mapping[member.name].get('original_capture_relative_path', member.name), member.size)

CONTAINERS.json maps members of redundant outer ZIP packages to their published loose counterparts or explicitly preserved nested variants. Container members were fully read and checked by CRC/SHA-256; matching loose source bytes were independently hashed. Scientific payloads are preserved while Finder/AppleDouble and original container owner metadata remain private. Byte-identical iterations may still overlap; this is not a benchmark split.

Privacy and verification

Personal workstation roots in selected JSON/text records become explicit archive-relative references or placeholders. PATHS.jsonl identifies each changed public copy and distinguishes its actual public hash/size from the historical original. Exact originals are privately retained and SHA-256 verified. JSON numerical values, array structure, media timing and scientific values are unchanged by these text edits. Historical embedded checksum claims continue to describe their original bytes; use the actual public manifest to check a privacy copy.

Legacy runtime DLLs and Finder metadata are retained privately, with original hashes and dispositions published separately. Runtime source and its licence are available in Diogenes-Archive.

All external JSON/JSONL/compressed metadata and technical text were fully read, as were JSON members inside depth ZIPs. All loose EXR and nested depth EXR headers, PNG/JPEG metadata, FLAC metadata, point-cloud/mesh structural metadata and relevant report images were checked. Representative visual review inspected every nonempty MP4’s first decoded frame plus selected scientific stills; it does not claim exhaustive viewing or audio transcription. Media samples show single-player campaign or validation scenes, without an identified private desktop/account overlay in those samples.

Publication is complete: all 9,204 public files (15,129,325,183 bytes) are preserved in 23 TAR shards. MANIFEST.jsonl lists every member’s actual size, SHA-256 and archive. ARCHIVE_INDEX.jsonl records each shard’s immutable upload revision and checksum. Every shard was checked against pinned Hugging Face LFS and anonymous range access. The corresponding original cloud objects, temporary public-copy objects and 4 redundant outer ZIP packages have been removed only after destination or private-original verification. All unique scientific payloads remain preserved; private originals and operating-system metadata remain private. Historical embedded hashes do not replace the manifest of the actual public bytes.

Read NOTICE.md for rights and historical restrictions. Tooling: Diogenes-Archive.

Downloads last month
104