Dataset Viewer
Duplicate
The dataset viewer is not available for this split.
Cannot extract the features (columns) for the split 'train' of the config 'default' of the dataset.
Error code:   FeaturesError
Exception:    ArrowInvalid
Message:      JSON parse error: Missing a name for object member. in row 0
Traceback:    Traceback (most recent call last):
                File "/usr/local/lib/python3.12/site-packages/datasets/packaged_modules/json/json.py", line 280, in _generate_tables
                  df = pandas_read_json(f)
                       ^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/packaged_modules/json/json.py", line 34, in pandas_read_json
                  return pd.read_json(path_or_buf, **kwargs)
                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/pandas/io/json/_json.py", line 815, in read_json
                  return json_reader.read()
                         ^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/pandas/io/json/_json.py", line 1014, in read
                  obj = self._get_object_parser(self.data)
                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/pandas/io/json/_json.py", line 1040, in _get_object_parser
                  obj = FrameParser(json, **kwargs).parse()
                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/pandas/io/json/_json.py", line 1176, in parse
                  self._parse()
                File "/usr/local/lib/python3.12/site-packages/pandas/io/json/_json.py", line 1391, in _parse
                  self.obj = DataFrame(
                             ^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/pandas/core/frame.py", line 778, in __init__
                  mgr = dict_to_mgr(data, index, columns, dtype=dtype, copy=copy, typ=manager)
                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/pandas/core/internals/construction.py", line 503, in dict_to_mgr
                  return arrays_to_mgr(arrays, columns, index, dtype=dtype, typ=typ, consolidate=copy)
                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/pandas/core/internals/construction.py", line 114, in arrays_to_mgr
                  index = _extract_index(arrays)
                          ^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/pandas/core/internals/construction.py", line 667, in _extract_index
                  raise ValueError("If using all scalar values, you must pass an index")
              ValueError: If using all scalar values, you must pass an index
              
              During handling of the above exception, another exception occurred:
              
              Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/split/first_rows.py", line 247, in compute_first_rows_from_streaming_response
                  iterable_dataset = iterable_dataset._resolve_features()
                                     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 4196, in _resolve_features
                  features = _infer_features_from_batch(self.with_format(None)._head())
                                                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 2533, in _head
                  return next(iter(self.iter(batch_size=n)))
                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 2711, in iter
                  for key, pa_table in ex_iterable.iter_arrow():
                                       ^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 2249, in _iter_arrow
                  yield from self.ex_iterable._iter_arrow()
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 494, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 384, in _iter_arrow
                  for key, pa_table in self.generate_tables_fn(**gen_kwags):
                                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/packaged_modules/json/json.py", line 283, in _generate_tables
                  raise e
                File "/usr/local/lib/python3.12/site-packages/datasets/packaged_modules/json/json.py", line 246, in _generate_tables
                  pa_table = paj.read_json(
                             ^^^^^^^^^^^^^^
                File "pyarrow/_json.pyx", line 342, in pyarrow._json.read_json
                File "pyarrow/error.pxi", line 155, in pyarrow.lib.pyarrow_internal_check_status
                File "pyarrow/error.pxi", line 92, in pyarrow.lib.check_status
              pyarrow.lib.ArrowInvalid: JSON parse error: Missing a name for object member. in row 0

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.

🎲 VSM Benchmarks β€” Cards & ChessImages

Two image benchmarks with very large, structured semantic spaces and fast, training-free validators, released with the paper "Score-Control for Hallucination Reduction in Diffusion Models" (VSM).

πŸ“„ Paper: https://arxiv.org/abs/2606.00377  Β·  πŸ’» Code: https://github.com/bhosalems/VSM

Both datasets are designed for systematic studies of hallucination in generative (especially diffusion) models. Each sample has a well-defined notion of validity, so a generated image can be checked β€” without any trained critic β€” for whether it lies on the data manifold (a legal board / a real card) or is a hallucination (an illegal configuration). This makes the hallucination rate a directly measurable quantity rather than a perceptual judgement.

Dataset Domain #Images Resolution Semantic space Validity check
Cards 2Γ—2 grids of playing cards 91,390 128Γ—128 ~10⁡ per-quadrant template match
ChessImages Chessboards rendered from FEN 196,062 256Γ—256 ~10⁴⁴ legal states FEN parse + python-chess legality

πŸ“¦ Repository layout

Image folders are shipped as .tar archives (a handful of large files uploads far faster than hundreds of thousands of tiny PNGs); labels and metadata are plain files.

mbhosale/VSM/
β”œβ”€β”€ ChessImages/
β”‚   β”œβ”€β”€ train_images.tar        # 176,938 Γ— 256Γ—256 PNG  -> train_images/
β”‚   β”œβ”€β”€ test_images.tar         #  19,124 Γ— 256Γ—256 PNG  -> test_images/
β”‚   β”œβ”€β”€ train_fen.json          # {"<image_id>": "<FEN>"}  (176,938 entries)
β”‚   └── test_fen.json           # {"<image_id>": "<FEN>"}  ( 19,124 entries)
β”œβ”€β”€ Cards/
β”‚   β”œβ”€β”€ card_imgs.tar           # 91,390 Γ— 128Γ—128 PNG  -> card_imgs/
β”‚   └── templates.tar           # 40 card-face templates (64Γ—64) for the validator
└── validators/                 # training-free hallucination checkers (see below)
    β”œβ”€β”€ chess_validator.py
    β”œβ”€β”€ cards_validator.py
    └── templates/chess/        # 26 chess-piece templates used by chess_validator.py

πŸƒ Cards

A Cards sample is a 2Γ—2 grid of four standard playing cards at 128Γ—128 (each quadrant is 64Γ—64). The card faces are drawn from 40 base cards β€” ranks A through 10 across the four suits (♣ ♦ β™₯ β™ ); no face cards (J/Q/K). With four independently drawn quadrants the semantic space is on the order of 10⁡.

  • card_imgs.tar β€” the full set of 91,390 grids.
  • templates.tar β€” the 40 single-card reference templates (resized to 64Γ—64), used by the training-free validator. Extracts to resized_templates_png/.

Hallucination check. Each of the four quadrants is template-matched (cv2.TM_CCOEFF_NORMED) against the 40 templates. If the worst quadrant match falls below a threshold, the image cannot be explained by four valid cards and is counted as a hallucination (wrong/garbled symbols, color/count errors, noise).

β™ŸοΈ ChessImages

A ChessImages sample is a 256Γ—256 rendering of a chessboard generated from a FEN string. The space of legal piece placements is astronomically large (~10⁴⁴), making this a stringent test of whether a generator stays on-manifold.

  • train_images.tar (176,938) and test_images.tar (19,124) β€” rendered boards; filenames are <image_id>.png.
  • train_fen.json / test_fen.json β€” map each <image_id> to its full FEN.

Hallucination check. The board image is parsed back to a FEN piece-placement string by template-matching each of the 64 squares, then its legality is tested with python-chess (chess.Board.is_valid()). Boards that violate the rules (two white kings, >8 pawns, pawns on the back rank, …) are counted as hallucinations. The validator can additionally report exact / fuzzy FEN-reconstruction accuracy against the ground-truth FENs (conditional setting).

πŸš€ Usage

Option A β€” via the VSM repo (recommended)

The VSM repo ships loaders and a downloader that fetch and unpack these archives for you:

python -m datasets.download --dataset chess --out ./data/ChessImages
python -m datasets.download --dataset cards --out ./data/Cards

Option B β€” directly with huggingface_hub

from huggingface_hub import snapshot_download
import tarfile, glob, os

root = snapshot_download(repo_id="mbhosale/VSM", repo_type="dataset",
                         allow_patterns=["ChessImages/*"])
for tar in glob.glob(os.path.join(root, "ChessImages", "**", "*.tar"), recursive=True):
    with tarfile.open(tar) as t:
        t.extractall(os.path.dirname(tar))
# -> ChessImages/train_images/*.png, test_images/*.png, train_fen.json, test_fen.json

βœ… Verifiers (training-free hallucination validators)

The two validators used in the paper are bundled here under validators/, so you can score generated images without cloning the full code repo. They need opencv-python, numpy, tqdm, and β€” for chess β€” python-chess.

ChessImages β€” parse each rendered board back to a FEN and test legality with python-chess; illegal boards count as hallucinations. The piece templates ship alongside the script, so no extra flags are required:

python validators/chess_validator.py --gen-dir /path/to/generated/boards
# optional: also report FEN-reconstruction accuracy against the ground truth
python validators/chess_validator.py --gen-dir /path/gen \
    --gt-json ChessImages/test_fen.json --conditional

Cards β€” template-match each of the 4 quadrants against the card templates; if the worst-matching quadrant falls below the threshold, the image is a hallucination:

tar -xf Cards/templates.tar                      # -> resized_templates_png/
python validators/cards_validator.py --gen-dir /path/to/generated/cards \
    --template-dir resized_templates_png

Each validator reports the hallucination rate (H%) per folder (repeat --gen-dir for multiple seeds). Full fidelity metrics (FID / CLIP-FID / FLD) live in the VSM repo under evaluation/.

βš–οΈ Intended use & license

These benchmarks are released for research on generative-model reliability and hallucination. They are synthetic renderings (rule-based card grids and chess positions) and contain no personal data. Released under CC BY 4.0 β€” please verify this matches your intended terms before redistribution, and cite the paper below.

πŸ“‘ Citation

@article{bhosale2026vsm,
  title   = {Score-Control for Hallucination Reduction in Diffusion Models},
  author  = {Bhosale, Mahesh and Devulapally, Naresh Kumar and Wasi, Abdul and
             Pham, Chau and Lokhande, Vishnu Suresh and Doermann, David},
  journal = {arXiv preprint arXiv:2606.00377},
  year    = {2026}
}

The ChessImages dataset uses python-chess and FEN strings sampled from VALUED.

Downloads last month
75

Papers for mbhosale/VSM