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/folder_based_builder/folder_based_builder.py", line 203, in _split_generators
                  for pa_metadata_table in self._read_metadata(downloaded_metadata_file, metadata_ext=metadata_ext):
                                           ~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/folder_based_builder/folder_based_builder.py", line 334, in _read_metadata
                  pa_table = paj.read_json(
                      io.BytesIO(batch), read_options=paj.ReadOptions(block_size=block_size)
                  )
                File "pyarrow/_json.pyx", line 364, 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
                  raise convert_status(status)
              pyarrow.lib.ArrowInvalid: JSON parse error: Column(/date) changed from string to number in row 11
              
              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.

Art-1k

1,000 public-domain two-dimensional artworks from five museum open-access programmes, chosen for breadth: paintings, prints, drawings and watercolours, miniatures and manuscript pages, from before 1500 to 1930, from Europe, the Americas, East Asia, South and West Asia and the Himalayas. Built for re-captioning and a text-to-image fine-tune. Not captioned yet.

Contents

  • data/: the 1,000 images as WebP, with metadata.jsonl.
  • reserve/: 813 more images that passed the same filters, same format, for replacements or for growing the set.
  • build/: BUILD.md (the full build log: queries, filters and drop counts, review rounds), quotas.json (target vs achieved for every quota) and dropped.jsonl (the 187 images removed in review, each with its reason and replacement).
  • scripts/: the recipe that built it, parameterised to extend it (see build/BUILD.md, "Extending") and to fetch the full-resolution originals.

Images. Each image is its 1024² aspect-ratio bucket: scaled to cover a bucket of about 1 megapixel with sides multiples of 16, then centre-cropped (on average 0.8 % of the area, at most 2.6 %), EXIF-transposed, converted to sRGB, and saved as WebP at quality 90. The source's original is at least 1024 px on its short side, with an aspect ratio between 0.4 and 2.5.

Rows. file_name, source, id, title, artist, date, begin_year, medium, classification, medium_category, region, period, subject, monochrome, japanese_print, width, height, bucket_width, bucket_height, crop_fraction, orig_width, orig_height, license, license_url, source_url (the museum's object page), original_url (the largest file the source serves), info_url, original_note (set when the source serves less than its master), curator_flag.

Originals. python3 -I scripts/fetch_originals.py data/metadata.jsonl OUT_DIR downloads them politely and resumably and checks their dimensions. The National Gallery serves at most 4096 px on the long side, the Art Institute 3000 px and the Rijksmuseum about 42 MP; Cleveland serves full TIFFs.

Sources and licences

Source Images Licence
National Gallery of Art, Washington 350 CC0 (open-access images)
Art Institute of Chicago 250 CC0 (public-domain works)
Cleveland Museum of Art 200 CC0
Rijksmuseum 100 Public Domain Mark 1.0
SMK, National Gallery of Denmark 100 Public Domain Mark 1.0

Per image: CC0 1.0 800, Public Domain Mark 1.0 200. Each row carries its licence URL and object page. The museums request, but do not require, a credit line (e.g. "Courtesy National Gallery of Art, Washington").

Composition

  • Medium: painting 492, print 224, drawing 197, other2d 87.
  • Region: Europe 600, Americas 184, East Asia 109, South/West Asia & Islamic 81, Other 26.
  • Period: 1500-1700 227, 1870-1930 215, 1800-1870 200, 1700-1800 195, pre-1500 163.
  • Subject: religious & mythological 208, portraits & figures 198, landscape & seascape 148, genre & everyday life 129, architecture & city views 69, allegory & other 61, animals & birds 61, still life & flowers 55, history & battle 55, botanical & natural history 16.
  • Colour: 250 monochrome or near-monochrome; 22 Japanese woodblock prints.

Subject, region and period come from keyword rules over the museum metadata and were corrected by visual review, so expect a few mislabels. Known gaps: no Korean art, little from Africa, Southeast Asia, Oceania or Latin America, few Persian or Ottoman works, no icons, few European manuscripts or posters.

How it was made

Candidates were drawn from each museum's open-access metadata, stratified by medium, region, period and subject, with each museum capped (35 / 25 / 20 / 20 / 10 %). About 4,100 were downloaded and filtered: nearly blank sheets, colour targets and rulers, near-duplicates (title and perceptual hash), and a visual review that removed photographs of objects, mostly-text pages, unfinished sketches, sheets with several works, ornament and illegible scans. The selected set then went through a review by Claude: one reviewer per 100 images, a critic of the whole set for repeated looks and gaps, and replacement from the reserve with each replacement looked at first. 187 images were swapped (repeated looks, works small inside large mounts or margins, duplicates, possible remaining copyright, weak scans, several works in one image) and about 370 metadata fields corrected.

Downloads last month
29