The dataset viewer is not available for this subset.
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/folder_based_builder/folder_based_builder.py", line 246, in _split_generators
raise ValueError(
"`file_name`, `*_file_name`, `file_names` or `*_file_names` must be present as dictionary key in metadata files"
)
ValueError: `file_name`, `*_file_name`, `file_names` or `*_file_names` must be present as dictionary key in metadata files
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.
wild3dbench — prompt suites and reference images
The evaluation inputs for wild3dbench, a reproducible benchmark for image-to-3D and text-to-3D generation.
Two suites, each usable from either referent — the prompt drives text-to-3D, the paired reference image drives image-to-3D, and because they describe the same object the two conditioning modes are directly comparable.
| Suite | Items | What it measures |
|---|---|---|
quality |
400 prompts, 400 images | Object quality: geometry plausibility and detail, texture quality, cross-view consistency, geometry–texture coherence. |
alignment |
100 prompts, 100 images, 395 atoms | Prompt alignment, scored as independent binary claims rather than one holistic number. |
Two files, one picture
Each prompt has exactly one reference image, distributed in two forms. images/ is the RGBA
master with a transparent background; images_grey/ is that same render flattened onto a light
grey backdrop, for pipelines that cannot accept an alpha channel. They are not two renders —
composite a master onto its own backdrop colour and it reproduces the flattened copy exactly.
So: 500 images, 1,000 files. Use whichever your pipeline wants, but use the same one throughout a run, and say which in your report — background handling is a real source of score variation in image-to-3D.
Why the atoms
Asking a judge "how well does this match the prompt, 1 to 5?" produces a number nobody can act
on. alignment/atoms.jsonl decomposes each prompt into the individual claims it makes:
{"id": "0001", "atoms": [
{"id": "a1", "type": "PA-Cat", "claim": "The object is a lamp."},
{"id": "a2", "type": "PA-Color", "claim": "The lamp is purple."},
{"id": "a3", "type": "PA-Shape", "claim": "The lamp's shade is cone-shaped."}]}
Each is judged pass / fail / unclear on its own, which turns one opaque score into a pass rate
per claim type — the difference between "alignment is 3.4" and "counting fails 60% of the time
while category succeeds 99%". PA-Cat is a hard gate: if the model made the wrong kind of
object, no other claim can rescue it.
Nine binding types (Color, Count, Part, Pose, Shape, Spatial, State, Style, Texture) are balanced at 32–33 claims each, so no single type dominates the aggregate.
Integrity
All 1,000 files are pinned by sha256 in checksums.json. The authoritative copy of that file ships
inside the wild3dbench package, not here — a mirror that could serve both the bytes and the
hashes that bless them would not be a check. The verifying download is:
pip install wild3dbench
wild3d suite download quality
wild3d suite download alignment
A checksum mismatch aborts and deletes the offending file rather than leaving it for the next
run to pick up. The hashes also feed Suite.fingerprint(), so a changed image changes the
fingerprint and stops the affected results from being pooled with results computed on the old
one.
Files
checksums.json sha256 of all 1000 files (browsing copy)
quality/metadata.jsonl 400 prompts, with category and modifier
quality/images/ 400 RGBA masters, <id>.png
quality/images_grey/ the same 400, flattened onto grey
alignment/metadata.jsonl 100 prompts, with atomicity and declared atom types
alignment/atoms.jsonl 395 authored claims, 3–5 per prompt
alignment/images/ 100 RGBA masters
alignment/images_grey/ the same 100, flattened onto grey
License and provenance
Prompts, atom records, and rubrics are original to this project, CC BY 4.0.
The reference images were synthesized with FLUX.2-dev and are redistributed under CC BY 4.0 on
the basis that the generating model's terms permit redistribution of its outputs. If you intend
to redistribute the images independently or use them commercially, satisfy yourself as to the
FLUX.2-dev output terms as they apply to your use. See LICENSE-DATA in the code repository for
the full statement.
Citation
See CITATION.cff in the code repository.
- Downloads last month
- 14