Dataset Viewer
Duplicate
The dataset viewer is not available for this split.
Cannot load the dataset split (in streaming mode) to extract the first rows.
Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
url: string
posters: int64
english_ui: bool
exact_download_checks: list<item: string>
  child 0, item: string
hash_navigation: bool
reconstruction_controls: bool
failed_mask_disclosed: bool
qwen_reference_loaded: bool
mobile_no_horizontal_overflow: bool
overview_thumbnails_loaded: int64
browser_errors: list<item: null>
  child 0, item: null
failed_http_responses: list<item: null>
  child 0, item: null
expected_asset_hosts: list<item: string>
  child 0, item: string
counting_notes: string
poster_counts: list<item: struct<sample_id: string, planned: int64, color_calls: int64, native_images: int64, deriv (... 31 chars omitted)
  child 0, item: struct<sample_id: string, planned: int64, color_calls: int64, native_images: int64, derived_rgba: in (... 19 chars omitted)
      child 0, sample_id: string
      child 1, planned: int64
      child 2, color_calls: int64
      child 3, native_images: int64
      child 4, derived_rgba: int64
      child 5, failed: int64
image_models_explicitly_reported: list<item: null>
  child 0, item: null
all_targets_attempted: bool
scope: string
failed_calls: list<item: struct<sample_id: string, layer_id: string, stage: string, notes: list<item: string>>>
  child 0, item: struct<sample_id: string, layer_id: string, stage: string, notes: list<item: string>>
      child 0, sample_id: string
      child 1, layer_id: string
      child 2, stage: string
      child 3, notes: list<item: string>
          child 0, item: string
pending: list<item: null>
  child 0, item: null
counts: struct<planned_layers: int64, color_calls: int64, recovered_results: int64, recorded_image_prompts:  (... 219 chars omitted)
  child 0, planned_layers: int64
  child 1, color_calls: int64
  child 2, recovered_results: int64
  child 3, recorded_image_prompts: int64
  child 4, native_images: int64
  child 5, backgrounds: int64
  child 6, native_opaque_foregrounds: int64
  child 7, opacity_calls: int64
  child 8, derived_rgba: int64
  child 9, resized_masks: int64
  child 10, near_empty_rgba: int64
  child 11, opacity_failures: int64
  child 12, color_failures: int64
to
{'posters': Value('int64'), 'all_targets_attempted': Value('bool'), 'counts': {'planned_layers': Value('int64'), 'color_calls': Value('int64'), 'recovered_results': Value('int64'), 'recorded_image_prompts': Value('int64'), 'native_images': Value('int64'), 'backgrounds': Value('int64'), 'native_opaque_foregrounds': Value('int64'), 'opacity_calls': Value('int64'), 'derived_rgba': Value('int64'), 'resized_masks': Value('int64'), 'near_empty_rgba': Value('int64'), 'opacity_failures': Value('int64'), 'color_failures': Value('int64')}, 'pending': List(Value('null')), 'failed_calls': List({'sample_id': Value('string'), 'layer_id': Value('string'), 'stage': Value('string'), 'notes': List(Value('string'))}), 'poster_counts': List({'sample_id': Value('string'), 'planned': Value('int64'), 'color_calls': Value('int64'), 'native_images': Value('int64'), 'derived_rgba': Value('int64'), 'failed': Value('int64')}), 'image_models_explicitly_reported': List(Value('null')), 'counting_notes': Value('string'), 'scope': Value('string')}
because column names don't match
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 147, in get_rows_or_raise
                  return get_rows(
                      dataset=dataset,
                  ...<4 lines>...
                      column_names=column_names,
                  )
                File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
                  return func(*args, **kwargs)
                File "/src/services/worker/src/worker/utils.py", line 127, in get_rows
                  rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
                File "/src/services/worker/src/worker/utils.py", line 483, in safe_iter
                  yield from ds.decode(False) if ds.features else ds
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2840, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2373, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2398, in _iter_arrow
                  for key, pa_table in self.ex_iterable._iter_arrow():
                                       ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 536, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 419, in _iter_arrow
                  for key, pa_table in self.generate_tables_fn(**gen_kwags):
                                       ~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 343, in _generate_tables
                  self._cast_table(pa_table, json_field_paths=json_field_paths),
                  ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
                  pa_table = table_cast(pa_table, self.info.features.arrow_schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2378, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2306, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              url: string
              posters: int64
              english_ui: bool
              exact_download_checks: list<item: string>
                child 0, item: string
              hash_navigation: bool
              reconstruction_controls: bool
              failed_mask_disclosed: bool
              qwen_reference_loaded: bool
              mobile_no_horizontal_overflow: bool
              overview_thumbnails_loaded: int64
              browser_errors: list<item: null>
                child 0, item: null
              failed_http_responses: list<item: null>
                child 0, item: null
              expected_asset_hosts: list<item: string>
                child 0, item: string
              counting_notes: string
              poster_counts: list<item: struct<sample_id: string, planned: int64, color_calls: int64, native_images: int64, deriv (... 31 chars omitted)
                child 0, item: struct<sample_id: string, planned: int64, color_calls: int64, native_images: int64, derived_rgba: in (... 19 chars omitted)
                    child 0, sample_id: string
                    child 1, planned: int64
                    child 2, color_calls: int64
                    child 3, native_images: int64
                    child 4, derived_rgba: int64
                    child 5, failed: int64
              image_models_explicitly_reported: list<item: null>
                child 0, item: null
              all_targets_attempted: bool
              scope: string
              failed_calls: list<item: struct<sample_id: string, layer_id: string, stage: string, notes: list<item: string>>>
                child 0, item: struct<sample_id: string, layer_id: string, stage: string, notes: list<item: string>>
                    child 0, sample_id: string
                    child 1, layer_id: string
                    child 2, stage: string
                    child 3, notes: list<item: string>
                        child 0, item: string
              pending: list<item: null>
                child 0, item: null
              counts: struct<planned_layers: int64, color_calls: int64, recovered_results: int64, recorded_image_prompts:  (... 219 chars omitted)
                child 0, planned_layers: int64
                child 1, color_calls: int64
                child 2, recovered_results: int64
                child 3, recorded_image_prompts: int64
                child 4, native_images: int64
                child 5, backgrounds: int64
                child 6, native_opaque_foregrounds: int64
                child 7, opacity_calls: int64
                child 8, derived_rgba: int64
                child 9, resized_masks: int64
                child 10, near_empty_rgba: int64
                child 11, opacity_failures: int64
                child 12, color_failures: int64
              to
              {'posters': Value('int64'), 'all_targets_attempted': Value('bool'), 'counts': {'planned_layers': Value('int64'), 'color_calls': Value('int64'), 'recovered_results': Value('int64'), 'recorded_image_prompts': Value('int64'), 'native_images': Value('int64'), 'backgrounds': Value('int64'), 'native_opaque_foregrounds': Value('int64'), 'opacity_calls': Value('int64'), 'derived_rgba': Value('int64'), 'resized_masks': Value('int64'), 'near_empty_rgba': Value('int64'), 'opacity_failures': Value('int64'), 'color_failures': Value('int64')}, 'pending': List(Value('null')), 'failed_calls': List({'sample_id': Value('string'), 'layer_id': Value('string'), 'stage': Value('string'), 'notes': List(Value('string'))}), 'poster_counts': List({'sample_id': Value('string'), 'planned': Value('int64'), 'color_calls': Value('int64'), 'native_images': Value('int64'), 'derived_rgba': Value('int64'), 'failed': Value('int64')}), 'image_models_explicitly_reported': List(Value('null')), 'counting_notes': Value('string'), 'scope': Value('string')}
              because column names don't match

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.

Codex poster layer baseline

Public source marketing posters, GPT-planned layer inventories, first image-tool outputs, GPT-generated opacity masks, and derived RGBA layers. The paired Space provides an English interactive inspector. These are model predictions, not ground-truth segmentation or original design assets.

Planning used GPT-6 Astra through codex exec. Image generation also ran through Codex's built-in image tool, which does not expose its exact backend model, snapshot, seed or quality settings. This dataset must not be attributed to a locked GPT Image 2 version.

Each foreground RGB output is conditioned on the original poster. If native transparency is absent, a separate image-tool call predicts a grayscale opacity mask from the generated RGB layer. Derived RGBA preserves RGB and copies grayscale mask values into alpha. Resizing, where required, is recorded per layer. No SAM, Qwen, OCR rerendering or manual edge cleanup is part of this branch.

Files

  • data/vi_###/original.*: unchanged source poster.
  • L##.png: first native color output, unchanged.
  • L##_matte.png: first GPT opacity-mask output, unchanged.
  • L##_rgba.png and .json: packed RGBA and conversion metadata.
  • BG.png: generated background with the planned foreground removed.
  • plan.json, prompts and result JSON: model planning, instructions and outcomes.
  • qwen.json: frozen metadata for the earlier Qwen candidate experiment; its native images remain in the separately linked previous Space.
  • source_preview.jpg and reconstruction_preview.png: resized overview thumbnails, separate from the native outputs.
  • pilot_probe/: the separate minimal native-transparency capability test.
  • reproduce/: Codex orchestration and channel-packing scripts.
  • protocol/artifact_validation.json: completeness and file-integrity checks, including failed model calls; these checks do not score semantic quality.
  • asset_manifest.json: file sizes and SHA-256 hashes.

The ZIP contains originals, native color outputs, masks, plans and metadata. It omits the duplicated derived RGBA files, which can be rebuilt from RGB and masks with python reproduce/pack_rgba.py --input-root data. All derived PNGs are individually available in the repository. Failed calls and poor outputs are retained rather than silently replaced. Generated hidden regions are plausible guesses, not verified recovery. The orchestration scripts retain the original workspace paths; adapt those paths and provide an authenticated Codex environment to rerun model calls. The hidden image backend is not version-locked, so generated pixels are not reproducible bit for bit. Channel packing from the included outputs is deterministic. After a quota interruption, 15 existing output files were recovered unchanged from their Codex session directories. Their result metadata is explicitly host-recovered; exact image prompts and tool-call counts are unavailable. The interruption and subsequent resume are recorded in protocol/quota_recovery.json.

Downloads last month
528

Space using Elfsong/codex-poster-layer-baseline-48 1