Complete camera-ready Croissant RAI provenance and publish checker validation

CROISSANT_VALIDATION_REPORT.txt

@@ -0,0 +1,1755 @@
+# CROISSANT VALIDATION REPORT
+================================================================================
+## VALIDATION RESULTS
+--------------------------------------------------------------------------------
+Starting validation for file: croissant.json
+### JSON Format Validation
+✓
+The file is valid JSON.
+### Croissant Schema Validation
+✓
+The dataset passes Croissant validation.
+### Responsible AI Metadata
+✓
+All required Responsible AI metadata fields are present.
+### Records Generation Test
+✓
+Record set 'entries' passed validation.
+Record set 'pairs' passed validation.
+Record set 'task3_edit_plans' passed validation.
+## JSON-LD REFERENCE
+================================================================================
+```json
+{
+  "@context": {
+    "@language": "en",
+    "@vocab": "https://schema.org/",
+    "citeAs": "cr:citeAs",
+    "column": "cr:column",
+    "conformsTo": "dct:conformsTo",
+    "cr": "http://mlcommons.org/croissant/",
+    "rai": "http://mlcommons.org/croissant/RAI/",
+    "data": {
+      "@id": "cr:data",
+      "@type": "@json"
+    },
+    "dataType": {
+      "@id": "cr:dataType",
+      "@type": "@vocab"
+    },
+    "dct": "http://purl.org/dc/terms/",
+    "equivalentProperty": "cr:equivalentProperty",
+    "examples": {
+      "@id": "cr:examples",
+      "@type": "@json"
+    },
+    "extract": "cr:extract",
+    "field": "cr:field",
+    "fileProperty": "cr:fileProperty",
+    "fileObject": "cr:fileObject",
+    "fileSet": "cr:fileSet",
+    "format": "cr:format",
+    "includes": "cr:includes",
+    "isLiveDataset": "cr:isLiveDataset",
+    "jsonPath": "cr:jsonPath",
+    "key": "cr:key",
+    "md5": "cr:md5",
+    "parentField": "cr:parentField",
+    "path": "cr:path",
+    "recordSet": "cr:recordSet",
+    "references": "cr:references",
+    "regex": "cr:regex",
+    "repeated": "cr:repeated",
+    "replace": "cr:replace",
+    "samplingRate": "cr:samplingRate",
+    "sc": "https://schema.org/",
+    "separator": "cr:separator",
+    "source": "cr:source",
+    "subField": "cr:subField",
+    "transform": "cr:transform",
+    "prov": "http://www.w3.org/ns/prov#",
+    "@base": "cr_base_iri/"
+  },
+  "@type": "sc:Dataset",
+  "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale",
+  "name": "GenScale",
+  "description": "Relative object scale benchmark for image generation and editing: 900 entries, 1643 pairwise scale records, and 100 precise editing plans. See README.md for the image assets and field conventions.",
+  "conformsTo": [
+    "http://mlcommons.org/croissant/1.0",
+    "http://mlcommons.org/croissant/RAI/1.0"
+  ],
+  "url": "https://huggingface.co/datasets/lingxiao2049/GenScale",
+  "license": "https://creativecommons.org/licenses/by/4.0/",
+  "version": "1.0.0",
+  "datePublished": "2026-10-17",
+  "inLanguage": [
+    "en"
+  ],
+  "isLiveDataset": false,
+  "creator": [
+    {
+      "@type": "sc:Person",
+      "name": "Lingxiao Li"
+    },
+    {
+      "@type": "sc:Person",
+      "name": "Max Whitton"
+    },
+    {
+      "@type": "sc:Person",
+      "name": "Ledell Wu"
+    },
+    {
+      "@type": "sc:Person",
+      "name": "Boqing Gong"
+    }
+  ],
+  "publisher": {
+    "@type": "sc:Person",
+    "name": "Lingxiao Li",
+    "url": "https://lingxiao-li.github.io/"
+  },
+  "citeAs": "@article{li2026genscale,\n  title={GenScale: A Benchmark for Relative Object Scale in Image Generation and Editing},\n  author={Li, Lingxiao and Whitton, Max and Wu, Ledell and Gong, Boqing},\n  journal={arXiv preprint arXiv:2609.00525},\n  year={2026},\n  url={https://arxiv.org/abs/2609.00525}\n}",
+  "keywords": [
+    "relative object scale",
+    "image generation",
+    "image editing",
+    "benchmark"
+  ],
+  "rai:dataCollection": "The GenScale research team curated a 100-category size knowledge base from COCO/LVIS category vocabularies and cited physical-size references; 97 non-human categories serve the common-object pool. ABO product dimensions and catalog reference images ground the human-product task. Model-assisted prompts and generated Task 1/2 images provide synthetic content, and 100 erroneous but editable generated scenes become Task 3 inputs. The release contains 433 unique image files (294 Task 2 reference images, 100 Task 3 generated inputs and 39 Task 3 references). It packages existing benchmark records, without new image generation or participant collection. Per-record collection dates/geographic sampling and numeric generation seeds are not retained; the released prompts, source task IDs, product IDs, dimension sources and preserved source images provide the available reconstruction inputs. See paper Sec. 3.1-3.2 and Appendix A.",
+  "rai:dataCollectionType": [
+    "Curated physical-size references and object vocabularies",
+    "Selected ABO catalog metadata and product reference images",
+    "Machine-assisted generation/editing prompts",
+    "Synthetic generated scenes selected for scale editing"
+  ],
+  "rai:dataCollectionMissingData": "Scenario-inapplicable fields are null or empty. Only selected benchmark reference/input images are included; original ABO candidate-pool paths require the upstream archive. All nonempty benchmark image/reference paths resolve within this release.",
+  "rai:dataCollectionRawData": "ABO: https://amazon-berkeley-objects.s3.amazonaws.com/index.html . Common-object metadata source_reference values are retained. This release includes 433 selected image files and does not redistribute the full original ABO archive.",
+  "rai:dataPreprocessingProtocol": [
+    "Absolute local paths converted to relative paths.",
+    "Frozen JSON flattened into entries, pairs, and task3_edit_plans CSV/Parquet tables using scripts/make_genscale_parquet.py.",
+    "No imputation of missing physical-size fields; original numerical annotations retained."
+  ],
+  "rai:dataAnnotationProtocol": [
+    "Benchmark target labels are physical characteristic-length A/B ratios and acceptable intervals derived from the curated size knowledge base or ABO dimensions with canonical human anchors. The precise-edit table adds the editable object, fixed reference object, resize direction and factor. These are not nine independent human annotations for every released relation.",
+    "The automatic evaluation protocol uses an ordinal 1-5 scale with 3 denoting plausible target scale; the judge uses anchor A and target B with the reciprocal B/A ratio convention. Nine human raters separately calibrated that judge using a custom annotation GUI; consensus uses the mode with median tie breaking. The paper describes the rubric, instructions and agreement analysis in Sec. 3.3 and Appendix B. Those private rater responses are not distributed as benchmark rows.",
+    "Task 3 selection requires a visible and editable scale error and removes missing/unidentifiable objects, duplicate/merged objects, severe artifacts, heavy occlusion/blur and extreme resizing cases. This is quality control for scorable/editable inputs, not an independent fairness or privacy audit."
+  ],
+  "rai:machineAnnotationTools": [
+    "Gemini API assistance is used for prompt construction/refinement and evaluation; the released refine_meta.gemini_model field records the refinement model where available.",
+    "The preserved Task 3 source provenance identifies FLUX.2 (61 unique inputs) and Qwen-Image-Edit-2511 (39 unique inputs). Source task IDs and prompts are released; exact dated API backends and per-image numeric RNG seeds are not fully retained.",
+    "Human judge-calibration annotations use the custom interface described in paper Appendix B. Public judging and aggregation implementations are in the GenScale code repository."
+  ],
+  "rai:dataUseCases": [
+    "Measure physically grounded pairwise relative object scale in image generation and editing, under the characteristic-length definitions and acceptable physical-size intervals in the released metadata.",
+    "Compare common-object generation (S1/S2), human-product generation with metric prompts (S3), and scale correction from selected erroneous generated inputs (S4/S5). The construct is pairwise relative scale, not full 3D geometry, image quality, authenticity, or all aspects of physical realism.",
+    "Evidence for this evaluative use is the paper's physical-size grounding, documented ordinal rubric, and calibration against human consensus (Sec. 3.1-3.3 and Appendix A-B). These establish a scoped evaluation protocol rather than universal measurement validity. Demographic fairness auditing, safety-critical measurement, authentication of images, and general-purpose model training have not been validated by this dataset."
+  ],
+  "rai:dataLimitations": [
+    "900 English-language prompt entries and 1643 pairwise records cover selected common objects and catalog products, not a representative sample of all objects, cultures, bodies or environments.",
+    "Physical metadata and canonical human-anchor lengths are reference assumptions; object sizes vary in the real world. Metadata dimensions can be imperfect and can conflict with prompt prose. Generated inputs can contain visual artifacts. See paper Sec. 3.1 and Appendix H.6.",
+    "S4 and S5 reuse the same 100 source images (61 common-object and 39 human-product cases); their 200 prompt entries are not 200 independent source images. Task 3 selects failed but editable generations and is not a random sample of generator outputs.",
+    "The Hugging Face train split name is a storage convention, not a prescribed model-training split. The 1000-product candidate metadata includes upstream ABO archive paths whose full raw image pool is not bundled. All nonempty benchmark image/reference paths resolve to the 433 packaged images.",
+    "Do not use this benchmark as a safety-critical physical measurement system, an authenticity detector, or evidence of complete 3D/physical correctness. Complete per-model generated outputs and private human annotation responses are outside this dataset package. Rescale weights are released separately at https://huggingface.co/lingxiao2049/Rescale under their own license."
+  ],
+  "rai:dataBiases": [
+    "Category selection starts from COCO/LVIS vocabularies and favors visually identifiable objects with relatively stable characteristic lengths. The resulting coverage is shaped by those vocabularies and by the authors' curation; ambiguous, highly variable and non-rigid categories are underrepresented.",
+    "Product examples inherit ABO catalog coverage, English listing-title selection, dimensional-data availability and a controlled characteristic-length range. This is not a global or demographic sample.",
+    "Fixed anthropometric anchors simplify variation in age, sex, disability and body size. Prompts and generated human depictions may reflect gender/age descriptors and source-model biases; no representative demographic labels or group-specific validity claims are provided.",
+    "Task 3 deliberately samples visible, scorable, editable scale errors from specified source-model outputs and applies visual quality filtering. Conclusions about correction should account for this selection. The automatic judge can inherit model biases; no independent fairness audit is claimed."
+  ],
+  "rai:personalSensitiveInformation": [
+    "The public fields describe objects, product titles/identifiers, dimensions, prompts and image paths. Images may show human faces/body parts and brands; prompt text can contain gender or age descriptors. These are not verified real-person demographic, identity or biometric labels.",
+    "Private rater identities, contact details, individual annotation forms, health records, political views and religious beliefs are not distributed in this package. No independent privacy audit or guarantee that all unstructured images are free of incidental personal information is claimed.",
+    "The paper describes generic human scale anchors, avoidance of named-person prompts, and annotation tasks that do not request sensitive participant disclosures (Appendix H.3-H.4). This metadata does not infer an IRB/exemption status or annotator demographic distribution."
+  ],
+  "rai:dataReleaseMaintenancePlan": "Report issues via https://github.com/lingxiao-li/GenScale/issues . Cite a dataset commit revision; checksums are supplied in SHA256SUMS. No fixed update cadence is promised.",
+  "distribution": [
+    {
+      "@type": "cr:FileObject",
+      "@id": "entries_csv",
+      "name": "data/genscale_entries.csv",
+      "description": "CSV table for GenScale entries. Empty cells denote absent fields.",
+      "contentUrl": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/3002a005586c105caf5b18c4ddd13db1be458625/data/genscale_entries.csv",
+      "encodingFormat": "text/csv",
+      "sha256": "8cf26356645a165d6bf64438322ab602e1a9d820fc76575e802916f4bae8a7f1",
+      "contentSize": "635212"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "pairs_csv",
+      "name": "data/genscale_pairs.csv",
+      "description": "CSV table for GenScale pairs. Empty cells denote absent fields.",
+      "contentUrl": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/3002a005586c105caf5b18c4ddd13db1be458625/data/genscale_pairs.csv",
+      "encodingFormat": "text/csv",
+      "sha256": "0f312e352e9f7cd20bbb6e63f97cd58bab1120f903ada97e2418a35417f26e6d",
+      "contentSize": "1088540"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "task3_edit_plans_csv",
+      "name": "data/genscale_task3_edit_plans.csv",
+      "description": "CSV table for GenScale task3_edit_plans. Empty cells denote absent fields.",
+      "contentUrl": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/3002a005586c105caf5b18c4ddd13db1be458625/data/genscale_task3_edit_plans.csv",
+      "encodingFormat": "text/csv",
+      "sha256": "0029aa2d231aed3a7d1b28f0cbd75a8a97e08f5498ff5a275f30731a19db5361",
+      "contentSize": "47682"
+    }
+  ],
+  "recordSet": [
+    {
+      "@type": "cr:RecordSet",
+      "@id": "entries",
+      "name": "entries",
+      "description": "GenScale entries table. See README.md for provenance and use.",
+      "key": [
+        {
+          "@id": "entries/entry_id"
+        }
+      ],
+      "field": [
+        {
+          "@type": "cr:Field",
+          "@id": "entries/entry_id",
+          "name": "entry_id",
+          "description": "entry id.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "entries_csv"
+            },
+            "extract": {
+              "column": "entry_id"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entries/task_id",
+          "name": "task_id",
+          "description": "task id.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "entries_csv"
+            },
+            "extract": {
+              "column": "task_id"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entries/task",
+          "name": "task",
+          "description": "task.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "entries_csv"
+            },
+            "extract": {
+              "column": "task"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entries/scenario",
+          "name": "scenario",
+          "description": "S1 natural depth, S2 extreme contrast, S3 human-product anchor, S4 discovery, or S5 precise editing.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "entries_csv"
+            },
+            "extract": {
+              "column": "scenario"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entries/source_task_id",
+          "name": "source_task_id",
+          "description": "source task id.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "entries_csv"
+            },
+            "extract": {
+              "column": "source_task_id"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entries/source_task_type",
+          "name": "source_task_type",
+          "description": "source task type.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "entries_csv"
+            },
+            "extract": {
+              "column": "source_task_type"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entries/prompt_type",
+          "name": "prompt_type",
+          "description": "prompt type.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "entries_csv"
+            },
+            "extract": {
+              "column": "prompt_type"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entries/num_objects",
+          "name": "num_objects",
+          "description": "num objects.",
+          "dataType": "sc:Integer",
+          "source": {
+            "fileObject": {
+              "@id": "entries_csv"
+            },
+            "extract": {
+              "column": "num_objects"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entries/num_pairs",
+          "name": "num_pairs",
+          "description": "num pairs.",
+          "dataType": "sc:Integer",
+          "source": {
+            "fileObject": {
+              "@id": "entries_csv"
+            },
+            "extract": {
+              "column": "num_pairs"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entries/objects_included",
+          "name": "objects_included",
+          "description": "Object names serialized as a JSON array string.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "entries_csv"
+            },
+            "extract": {
+              "column": "objects_included"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entries/prompt",
+          "name": "prompt",
+          "description": "Generation or editing instruction in English.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "entries_csv"
+            },
+            "extract": {
+              "column": "prompt"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entries/reference_image_path",
+          "name": "reference_image_path",
+          "description": "Reference image path relative to dataset root, where available.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "entries_csv"
+            },
+            "extract": {
+              "column": "reference_image_path"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entries/image_path",
+          "name": "image_path",
+          "description": "Input image path relative to dataset root; empty for tasks without an input scene.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "entries_csv"
+            },
+            "extract": {
+              "column": "image_path"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entries/raw_listing_title",
+          "name": "raw_listing_title",
+          "description": "raw listing title.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "entries_csv"
+            },
+            "extract": {
+              "column": "raw_listing_title"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entries/source_size_score",
+          "name": "source_size_score",
+          "description": "Original five-point scale score for the selected source input; 3 denotes plausible.",
+          "dataType": "sc:Float",
+          "source": {
+            "fileObject": {
+              "@id": "entries_csv"
+            },
+            "extract": {
+              "column": "source_size_score"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entries/pair_key",
+          "name": "pair_key",
+          "description": "pair key.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "entries_csv"
+            },
+            "extract": {
+              "column": "pair_key"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entries/generated_ratio",
+          "name": "generated_ratio",
+          "description": "generated ratio.",
+          "dataType": "sc:Float",
+          "source": {
+            "fileObject": {
+              "@id": "entries_csv"
+            },
+            "extract": {
+              "column": "generated_ratio"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entries/target_ratio",
+          "name": "target_ratio",
+          "description": "target ratio.",
+          "dataType": "sc:Float",
+          "source": {
+            "fileObject": {
+              "@id": "entries_csv"
+            },
+            "extract": {
+              "column": "target_ratio"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entries/product_scale.dataset_source",
+          "name": "product_scale.dataset_source",
+          "description": "product scale: dataset source.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "entries_csv"
+            },
+            "extract": {
+              "column": "product_scale.dataset_source"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entries/product_scale.category_id",
+          "name": "product_scale.category_id",
+          "description": "product scale: category id.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "entries_csv"
+            },
+            "extract": {
+              "column": "product_scale.category_id"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entries/product_scale.category_name",
+          "name": "product_scale.category_name",
+          "description": "product scale: category name.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "entries_csv"
+            },
+            "extract": {
+              "column": "product_scale.category_name"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entries/product_scale.source_reference",
+          "name": "product_scale.source_reference",
+          "description": "product scale: source reference.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "entries_csv"
+            },
+            "extract": {
+              "column": "product_scale.source_reference"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entries/product_scale.length_cm",
+          "name": "product_scale.length_cm",
+          "description": "product scale: length cm.",
+          "dataType": "sc:Float",
+          "source": {
+            "fileObject": {
+              "@id": "entries_csv"
+            },
+            "extract": {
+              "column": "product_scale.length_cm"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entries/product_scale.width_cm",
+          "name": "product_scale.width_cm",
+          "description": "product scale: width cm.",
+          "dataType": "sc:Float",
+          "source": {
+            "fileObject": {
+              "@id": "entries_csv"
+            },
+            "extract": {
+              "column": "product_scale.width_cm"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entries/product_scale.height_cm",
+          "name": "product_scale.height_cm",
+          "description": "product scale: height cm.",
+          "dataType": "sc:Float",
+          "source": {
+            "fileObject": {
+              "@id": "entries_csv"
+            },
+            "extract": {
+              "column": "product_scale.height_cm"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entries/product_scale.typical_len_cm",
+          "name": "product_scale.typical_len_cm",
+          "description": "product scale: typical len cm.",
+          "dataType": "sc:Float",
+          "source": {
+            "fileObject": {
+              "@id": "entries_csv"
+            },
+            "extract": {
+              "column": "product_scale.typical_len_cm"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entries/product_scale.min_len_cm",
+          "name": "product_scale.min_len_cm",
+          "description": "product scale: min len cm.",
+          "dataType": "sc:Float",
+          "source": {
+            "fileObject": {
+              "@id": "entries_csv"
+            },
+            "extract": {
+              "column": "product_scale.min_len_cm"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entries/product_scale.max_len_cm",
+          "name": "product_scale.max_len_cm",
+          "description": "product scale: max len cm.",
+          "dataType": "sc:Float",
+          "source": {
+            "fileObject": {
+              "@id": "entries_csv"
+            },
+            "extract": {
+              "column": "product_scale.max_len_cm"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entries/refine_meta.refined_with",
+          "name": "refine_meta.refined_with",
+          "description": "refine meta: refined with.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "entries_csv"
+            },
+            "extract": {
+              "column": "refine_meta.refined_with"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entries/refine_meta.gemini_model",
+          "name": "refine_meta.gemini_model",
+          "description": "refine meta: gemini model.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "entries_csv"
+            },
+            "extract": {
+              "column": "refine_meta.gemini_model"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entries/edit_plan.edit_target_object",
+          "name": "edit_plan.edit_target_object",
+          "description": "edit plan: edit target object.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "entries_csv"
+            },
+            "extract": {
+              "column": "edit_plan.edit_target_object"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entries/edit_plan.reference_object",
+          "name": "edit_plan.reference_object",
+          "description": "edit plan: reference object.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "entries_csv"
+            },
+            "extract": {
+              "column": "edit_plan.reference_object"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entries/edit_plan.scale_factor",
+          "name": "edit_plan.scale_factor",
+          "description": "edit plan: scale factor.",
+          "dataType": "sc:Float",
+          "source": {
+            "fileObject": {
+              "@id": "entries_csv"
+            },
+            "extract": {
+              "column": "edit_plan.scale_factor"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entries/edit_plan.direction",
+          "name": "edit_plan.direction",
+          "description": "edit plan: direction.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "entries_csv"
+            },
+            "extract": {
+              "column": "edit_plan.direction"
+            }
+          }
+        }
+      ]
+    },
+    {
+      "@type": "cr:RecordSet",
+      "@id": "pairs",
+      "name": "pairs",
+      "description": "GenScale pairs table. See README.md for provenance and use.",
+      "key": [
+        {
+          "@id": "pairs/pair_id"
+        }
+      ],
+      "field": [
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/pair_id",
+          "name": "pair_id",
+          "description": "pair id.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "pairs_csv"
+            },
+            "extract": {
+              "column": "pair_id"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/entry_id",
+          "name": "entry_id",
+          "description": "entry id.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "pairs_csv"
+            },
+            "extract": {
+              "column": "entry_id"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/task_id",
+          "name": "task_id",
+          "description": "task id.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "pairs_csv"
+            },
+            "extract": {
+              "column": "task_id"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/task",
+          "name": "task",
+          "description": "task.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "pairs_csv"
+            },
+            "extract": {
+              "column": "task"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/scenario",
+          "name": "scenario",
+          "description": "S1 natural depth, S2 extreme contrast, S3 human-product anchor, S4 discovery, or S5 precise editing.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "pairs_csv"
+            },
+            "extract": {
+              "column": "scenario"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/source_task_id",
+          "name": "source_task_id",
+          "description": "source task id.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "pairs_csv"
+            },
+            "extract": {
+              "column": "source_task_id"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/source_task_type",
+          "name": "source_task_type",
+          "description": "source task type.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "pairs_csv"
+            },
+            "extract": {
+              "column": "source_task_type"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/prompt_type",
+          "name": "prompt_type",
+          "description": "prompt type.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "pairs_csv"
+            },
+            "extract": {
+              "column": "prompt_type"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/pair_key",
+          "name": "pair_key",
+          "description": "pair key.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "pairs_csv"
+            },
+            "extract": {
+              "column": "pair_key"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/object_a",
+          "name": "object_a",
+          "description": "object a.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "pairs_csv"
+            },
+            "extract": {
+              "column": "object_a"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/object_b",
+          "name": "object_b",
+          "description": "object b.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "pairs_csv"
+            },
+            "extract": {
+              "column": "object_b"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/target_ratio_a_to_b",
+          "name": "target_ratio_a_to_b",
+          "description": "Intended characteristic physical length(A) / length(B). Judge code uses the reciprocal B/A convention.",
+          "dataType": "sc:Float",
+          "source": {
+            "fileObject": {
+              "@id": "pairs_csv"
+            },
+            "extract": {
+              "column": "target_ratio_a_to_b"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/acceptable_ratio_min",
+          "name": "acceptable_ratio_min",
+          "description": "Lower acceptable physical A/B ratio.",
+          "dataType": "sc:Float",
+          "source": {
+            "fileObject": {
+              "@id": "pairs_csv"
+            },
+            "extract": {
+              "column": "acceptable_ratio_min"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/acceptable_ratio_max",
+          "name": "acceptable_ratio_max",
+          "description": "Upper acceptable physical A/B ratio.",
+          "dataType": "sc:Float",
+          "source": {
+            "fileObject": {
+              "@id": "pairs_csv"
+            },
+            "extract": {
+              "column": "acceptable_ratio_max"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/generated_ratio_a_to_b",
+          "name": "generated_ratio_a_to_b",
+          "description": "Estimated generated-image A/B ratio used in constructing the precise editing instruction, when provided.",
+          "dataType": "sc:Float",
+          "source": {
+            "fileObject": {
+              "@id": "pairs_csv"
+            },
+            "extract": {
+              "column": "generated_ratio_a_to_b"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/source_size_score",
+          "name": "source_size_score",
+          "description": "Original five-point scale score for the selected source input; 3 denotes plausible.",
+          "dataType": "sc:Float",
+          "source": {
+            "fileObject": {
+              "@id": "pairs_csv"
+            },
+            "extract": {
+              "column": "source_size_score"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/objects_included",
+          "name": "objects_included",
+          "description": "Object names serialized as a JSON array string.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "pairs_csv"
+            },
+            "extract": {
+              "column": "objects_included"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/prompt",
+          "name": "prompt",
+          "description": "Generation or editing instruction in English.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "pairs_csv"
+            },
+            "extract": {
+              "column": "prompt"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/reference_image_path",
+          "name": "reference_image_path",
+          "description": "Reference image path relative to dataset root, where available.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "pairs_csv"
+            },
+            "extract": {
+              "column": "reference_image_path"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/image_path",
+          "name": "image_path",
+          "description": "Input image path relative to dataset root; empty for tasks without an input scene.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "pairs_csv"
+            },
+            "extract": {
+              "column": "image_path"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/product_scale.dataset_source",
+          "name": "product_scale.dataset_source",
+          "description": "product scale: dataset source.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "pairs_csv"
+            },
+            "extract": {
+              "column": "product_scale.dataset_source"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/product_scale.category_id",
+          "name": "product_scale.category_id",
+          "description": "product scale: category id.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "pairs_csv"
+            },
+            "extract": {
+              "column": "product_scale.category_id"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/product_scale.category_name",
+          "name": "product_scale.category_name",
+          "description": "product scale: category name.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "pairs_csv"
+            },
+            "extract": {
+              "column": "product_scale.category_name"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/product_scale.source_reference",
+          "name": "product_scale.source_reference",
+          "description": "product scale: source reference.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "pairs_csv"
+            },
+            "extract": {
+              "column": "product_scale.source_reference"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/product_scale.length_cm",
+          "name": "product_scale.length_cm",
+          "description": "product scale: length cm.",
+          "dataType": "sc:Float",
+          "source": {
+            "fileObject": {
+              "@id": "pairs_csv"
+            },
+            "extract": {
+              "column": "product_scale.length_cm"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/product_scale.width_cm",
+          "name": "product_scale.width_cm",
+          "description": "product scale: width cm.",
+          "dataType": "sc:Float",
+          "source": {
+            "fileObject": {
+              "@id": "pairs_csv"
+            },
+            "extract": {
+              "column": "product_scale.width_cm"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/product_scale.height_cm",
+          "name": "product_scale.height_cm",
+          "description": "product scale: height cm.",
+          "dataType": "sc:Float",
+          "source": {
+            "fileObject": {
+              "@id": "pairs_csv"
+            },
+            "extract": {
+              "column": "product_scale.height_cm"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/product_scale.typical_len_cm",
+          "name": "product_scale.typical_len_cm",
+          "description": "product scale: typical len cm.",
+          "dataType": "sc:Float",
+          "source": {
+            "fileObject": {
+              "@id": "pairs_csv"
+            },
+            "extract": {
+              "column": "product_scale.typical_len_cm"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/product_scale.min_len_cm",
+          "name": "product_scale.min_len_cm",
+          "description": "product scale: min len cm.",
+          "dataType": "sc:Float",
+          "source": {
+            "fileObject": {
+              "@id": "pairs_csv"
+            },
+            "extract": {
+              "column": "product_scale.min_len_cm"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/product_scale.max_len_cm",
+          "name": "product_scale.max_len_cm",
+          "description": "product scale: max len cm.",
+          "dataType": "sc:Float",
+          "source": {
+            "fileObject": {
+              "@id": "pairs_csv"
+            },
+            "extract": {
+              "column": "product_scale.max_len_cm"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/edit_plan.edit_target_object",
+          "name": "edit_plan.edit_target_object",
+          "description": "edit plan: edit target object.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "pairs_csv"
+            },
+            "extract": {
+              "column": "edit_plan.edit_target_object"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/edit_plan.reference_object",
+          "name": "edit_plan.reference_object",
+          "description": "edit plan: reference object.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "pairs_csv"
+            },
+            "extract": {
+              "column": "edit_plan.reference_object"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/edit_plan.scale_factor",
+          "name": "edit_plan.scale_factor",
+          "description": "edit plan: scale factor.",
+          "dataType": "sc:Float",
+          "source": {
+            "fileObject": {
+              "@id": "pairs_csv"
+            },
+            "extract": {
+              "column": "edit_plan.scale_factor"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/edit_plan.direction",
+          "name": "edit_plan.direction",
+          "description": "edit plan: direction.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "pairs_csv"
+            },
+            "extract": {
+              "column": "edit_plan.direction"
+            }
+          }
+        }
+      ]
+    },
+    {
+      "@type": "cr:RecordSet",
+      "@id": "task3_edit_plans",
+      "name": "task3_edit_plans",
+      "description": "GenScale task3_edit_plans table. See README.md for provenance and use.",
+      "key": [
+        {
+          "@id": "task3_edit_plans/entry_id"
+        }
+      ],
+      "field": [
+        {
+          "@type": "cr:Field",
+          "@id": "task3_edit_plans/entry_id",
+          "name": "entry_id",
+          "description": "entry id.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "task3_edit_plans_csv"
+            },
+            "extract": {
+              "column": "entry_id"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "task3_edit_plans/task_id",
+          "name": "task_id",
+          "description": "task id.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "task3_edit_plans_csv"
+            },
+            "extract": {
+              "column": "task_id"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "task3_edit_plans/scenario",
+          "name": "scenario",
+          "description": "S1 natural depth, S2 extreme contrast, S3 human-product anchor, S4 discovery, or S5 precise editing.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "task3_edit_plans_csv"
+            },
+            "extract": {
+              "column": "scenario"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "task3_edit_plans/source_task_id",
+          "name": "source_task_id",
+          "description": "source task id.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "task3_edit_plans_csv"
+            },
+            "extract": {
+              "column": "source_task_id"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "task3_edit_plans/source_task_type",
+          "name": "source_task_type",
+          "description": "source task type.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "task3_edit_plans_csv"
+            },
+            "extract": {
+              "column": "source_task_type"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "task3_edit_plans/prompt_type",
+          "name": "prompt_type",
+          "description": "prompt type.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "task3_edit_plans_csv"
+            },
+            "extract": {
+              "column": "prompt_type"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "task3_edit_plans/pair_key",
+          "name": "pair_key",
+          "description": "pair key.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "task3_edit_plans_csv"
+            },
+            "extract": {
+              "column": "pair_key"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "task3_edit_plans/object_a",
+          "name": "object_a",
+          "description": "object a.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "task3_edit_plans_csv"
+            },
+            "extract": {
+              "column": "object_a"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "task3_edit_plans/object_b",
+          "name": "object_b",
+          "description": "object b.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "task3_edit_plans_csv"
+            },
+            "extract": {
+              "column": "object_b"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "task3_edit_plans/generated_ratio_a_to_b",
+          "name": "generated_ratio_a_to_b",
+          "description": "Estimated generated-image A/B ratio used in constructing the precise editing instruction, when provided.",
+          "dataType": "sc:Float",
+          "source": {
+            "fileObject": {
+              "@id": "task3_edit_plans_csv"
+            },
+            "extract": {
+              "column": "generated_ratio_a_to_b"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "task3_edit_plans/target_ratio_a_to_b",
+          "name": "target_ratio_a_to_b",
+          "description": "Intended characteristic physical length(A) / length(B). Judge code uses the reciprocal B/A convention.",
+          "dataType": "sc:Float",
+          "source": {
+            "fileObject": {
+              "@id": "task3_edit_plans_csv"
+            },
+            "extract": {
+              "column": "target_ratio_a_to_b"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "task3_edit_plans/source_size_score",
+          "name": "source_size_score",
+          "description": "Original five-point scale score for the selected source input; 3 denotes plausible.",
+          "dataType": "sc:Integer",
+          "source": {
+            "fileObject": {
+              "@id": "task3_edit_plans_csv"
+            },
+            "extract": {
+              "column": "source_size_score"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "task3_edit_plans/edit_target_object",
+          "name": "edit_target_object",
+          "description": "edit target object.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "task3_edit_plans_csv"
+            },
+            "extract": {
+              "column": "edit_target_object"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "task3_edit_plans/reference_object",
+          "name": "reference_object",
+          "description": "reference object.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "task3_edit_plans_csv"
+            },
+            "extract": {
+              "column": "reference_object"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "task3_edit_plans/scale_factor",
+          "name": "scale_factor",
+          "description": "scale factor.",
+          "dataType": "sc:Float",
+          "source": {
+            "fileObject": {
+              "@id": "task3_edit_plans_csv"
+            },
+            "extract": {
+              "column": "scale_factor"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "task3_edit_plans/direction",
+          "name": "direction",
+          "description": "direction.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "task3_edit_plans_csv"
+            },
+            "extract": {
+              "column": "direction"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "task3_edit_plans/prompt",
+          "name": "prompt",
+          "description": "Generation or editing instruction in English.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "task3_edit_plans_csv"
+            },
+            "extract": {
+              "column": "prompt"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "task3_edit_plans/image_path",
+          "name": "image_path",
+          "description": "Input image path relative to dataset root; empty for tasks without an input scene.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "task3_edit_plans_csv"
+            },
+            "extract": {
+              "column": "image_path"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "task3_edit_plans/reference_image_path",
+          "name": "reference_image_path",
+          "description": "Reference image path relative to dataset root, where available.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "task3_edit_plans_csv"
+            },
+            "extract": {
+              "column": "reference_image_path"
+            }
+          }
+        }
+      ]
+    }
+  ],
+  "dateModified": "2026-10-17",
+  "rai:hasSyntheticData": true,
+  "rai:dataSocialImpact": "Positive potential: make relative-scale failures in image generation and editing measurable, support reproducible evaluation, and inform more physically plausible product visualization, design tools and simulation. These are potential uses, not independently established deployment outcomes. Negative potential: correcting scale artifacts can make manipulated or synthetic images more convincing and enable misleading product depictions or other deceptive visual communication. The selected English object/product vocabulary and fixed body-size anchors may also encourage overgeneralization to underrepresented people, cultures or physical environments. Mitigations documented in the release include source/task identifiers, explicit synthetic-data disclosure, published limitations, separate asset licenses, and exclusion of private rater records and credentials. The paper recommends research use and disclosure of synthetic editing, rather than treating scale plausibility as proof of authenticity or overall physical correctness. No independent fairness, privacy or misuse audit is claimed. See the paper Appendix H.3 and H.5-H.6.",
+  "prov:wasDerivedFrom": [
+    {
+      "@id": "https://amazon-berkeley-objects.s3.amazonaws.com/index.html",
+      "@type": "prov:Entity",
+      "prov:label": "Amazon Berkeley Objects (ABO)",
+      "prov:description": "Product identifiers, dimensions and selected catalog reference images; original source assets remain subject to ABO attribution and CC BY 4.0."
+    },
+    {
+      "@id": "https://cocodataset.org/",
+      "@type": "prov:Entity",
+      "prov:label": "COCO category vocabulary",
+      "prov:description": "Category names inform the curated object vocabulary. COCO images and complete annotations are not redistributed as this benchmark."
+    },
+    {
+      "@id": "https://www.lvisdataset.org/",
+      "@type": "prov:Entity",
+      "prov:label": "LVIS category vocabulary",
+      "prov:description": "Category names inform the curated object vocabulary. LVIS images and complete annotations are not redistributed as this benchmark."
+    },
+    {
+      "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/3002a005586c105caf5b18c4ddd13db1be458625/metadata/authoritative_kb_3d_100.csv",
+      "@type": "prov:Entity",
+      "prov:label": "GenScale physical-size construction seed table",
+      "prov:description": "100 category-level dimension records, including per-row source_reference citations for anthropometric, biological/food, manufactured-object, traffic and sports-size references."
+    },
+    {
+      "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/3002a005586c105caf5b18c4ddd13db1be458625/metadata/abo_local_sampled_1000_representative_anonymized.csv",
+      "@type": "prov:Entity",
+      "prov:label": "ABO candidate construction seed table",
+      "prov:description": "1000 candidate products and dimensions used in benchmark construction; upstream raw image paths are identifiers for the ABO archive."
+    },
+    {
+      "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/3002a005586c105caf5b18c4ddd13db1be458625/GenScale_Benchmark.json",
+      "@type": "prov:Entity",
+      "prov:label": "Frozen prompt and source-task seed records",
+      "prov:description": "The preserved generation/editing prompts, task identifiers, source task identifiers and physical-ratio metadata reconstruct the available synthetic-data inputs; exact per-image numeric random seeds are not supplied."
+    }
+  ],
+  "prov:wasGeneratedBy": [
+    {
+      "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale#activity-vocabulary-and-sizes",
+      "@type": "prov:Activity",
+      "prov:type": "collection",
+      "prov:label": "Curate object vocabulary and physical-size priors",
+      "prov:description": "Research-team curation of visually identifiable categories with stable physical extent, using cited size references and anthropometric anchors. Source vocabularies and per-row dimension references are preserved; collection timestamps and a geographic sampling frame are not available. Paper Sec. 3.1 and Appendix A.1.",
+      "prov:used": [
+        {
+          "@id": "https://cocodataset.org/"
+        },
+        {
+          "@id": "https://www.lvisdataset.org/"
+        },
+        {
+          "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/3002a005586c105caf5b18c4ddd13db1be458625/metadata/authoritative_kb_3d_100.csv"
+        }
+      ],
+      "prov:wasAssociatedWith": [
+        {
+          "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale#genscale-research-team",
+          "@type": "prov:Agent",
+          "prov:label": "GenScale authors / benchmark research team"
+        }
+      ]
+    },
+    {
+      "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale#activity-product-filtering",
+      "@type": "prov:Activity",
+      "prov:type": "preprocessing",
+      "prov:label": "Filter ABO products and assign human anchors",
+      "prov:description": "Require an English title, valid dimensions and a retrievable main image; convert dimensions to centimeters and use the maximum axis as characteristic length. The paper retains characteristic lengths in [5,250] cm, then assigns hand, face/head, foot/leg or full-body anchors by the documented size bands. Product identifiers and dimensions remain in the released records. Paper Appendix A.3.",
+      "prov:used": [
+        {
+          "@id": "https://amazon-berkeley-objects.s3.amazonaws.com/index.html"
+        },
+        {
+          "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/3002a005586c105caf5b18c4ddd13db1be458625/metadata/abo_local_sampled_1000_representative_anonymized.csv"
+        }
+      ],
+      "prov:wasAssociatedWith": [
+        {
+          "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale#genscale-research-team",
+          "@type": "prov:Agent",
+          "prov:label": "GenScale authors / benchmark research team"
+        }
+      ]
+    },
+    {
+      "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale#activity-prompt-construction",
+      "@type": "prov:Activity",
+      "prov:type": "collection",
+      "prov:label": "Construct machine-assisted benchmark prompts",
+      "prov:description": "Generate common-object prompts and product/human interaction prompts from selected objects, product metadata and dimensions using Gemini-assisted construction/refinement. The frozen prompt text and available refine_meta model identifier are released. Numeric RNG seeds are not documented per prompt. Paper Sec. 3.2 and Appendix A.",
+      "prov:used": [
+        {
+          "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/3002a005586c105caf5b18c4ddd13db1be458625/metadata/authoritative_kb_3d_100.csv"
+        },
+        {
+          "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/3002a005586c105caf5b18c4ddd13db1be458625/metadata/abo_local_sampled_1000_representative_anonymized.csv"
+        },
+        {
+          "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/3002a005586c105caf5b18c4ddd13db1be458625/GenScale_Benchmark.json"
+        }
+      ],
+      "prov:wasAssociatedWith": [
+        {
+          "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale#genscale-research-team",
+          "@type": "prov:Agent",
+          "prov:label": "GenScale authors / benchmark research team"
+        },
+        {
+          "@id": "https://ai.google.dev/gemini-api/docs",
+          "@type": "prov:SoftwareAgent",
+          "prov:label": "Gemini API construction/refinement tools"
+        }
+      ]
+    },
+    {
+      "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale#activity-synthetic-source-images",
+      "@type": "prov:Activity",
+      "prov:type": "collection",
+      "prov:label": "Generate and select Task 3 synthetic image inputs",
+      "prov:description": "The preserved source records identify 61 FLUX.2 common-object images and 39 Qwen-Image-Edit-2511 human-product images. Select failed but scorable/editable pairwise cases, apply the documented object-visibility/artifact/correction-factor quality controls, and create S4/S5 prompts for each of the same 100 source images. Exact API backend revisions and image-level numeric seeds are not fully retained; the images and their source-task prompts/IDs are preserved. Paper Sec. 3.2 and Appendix A.4-A.5.",
+      "prov:used": [
+        {
+          "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/3002a005586c105caf5b18c4ddd13db1be458625/GenScale_Benchmark.json"
+        }
+      ],
+      "prov:wasAssociatedWith": [
+        {
+          "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale#genscale-research-team",
+          "@type": "prov:Agent",
+          "prov:label": "GenScale authors / benchmark research team"
+        },
+        {
+          "@id": "https://bfl.ai/blog/flux-2",
+          "@type": "prov:SoftwareAgent",
+          "prov:label": "FLUX.2 source image generator"
+        },
+        {
+          "@id": "https://huggingface.co/Qwen/Qwen-Image-Edit-2511",
+          "@type": "prov:SoftwareAgent",
+          "prov:label": "Qwen-Image-Edit-2511 source image generator"
+        }
+      ]
+    },
+    {
+      "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale#activity-ratio-and-edit-labels",
+      "@type": "prov:Activity",
+      "prov:type": "annotation",
+      "prov:label": "Derive physical-ratio labels and precise edit plans",
+      "prov:description": "Compute intended physical A/B ratios and acceptable intervals from characteristic sizes. For precise editing, store the target and reference object, rendered ratio, direction and multiplicative resize factor in the released edit-plan records. Separate human judge calibration uses nine raters with the five-point rubric and mode/median consensus described in Sec. 3.3 and Appendix B; it is not nine human labels per benchmark relation. Private annotation forms/rater identities are not part of these tables, and no new demographic or ethics-review claims are made here.",
+      "prov:used": [
+        {
+          "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/3002a005586c105caf5b18c4ddd13db1be458625/metadata/authoritative_kb_3d_100.csv"
+        },
+        {
+          "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/3002a005586c105caf5b18c4ddd13db1be458625/metadata/abo_local_sampled_1000_representative_anonymized.csv"
+        },
+        {
+          "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/3002a005586c105caf5b18c4ddd13db1be458625/GenScale_Benchmark.json"
+        },
+        {
+          "@id": "https://arxiv.org/abs/2609.00525"
+        }
+      ],
+      "prov:wasAssociatedWith": [
+        {
+          "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale#genscale-research-team",
+          "@type": "prov:Agent",
+          "prov:label": "GenScale authors / benchmark research team"
+        },
+        {
+          "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale#calibration-raters",
+          "@type": "prov:Agent",
+          "prov:label": "Nine judge-calibration raters; identities not distributed"
+        }
+      ]
+    },
+    {
+      "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale#activity-public-packaging",
+      "@type": "prov:Activity",
+      "prov:type": "preprocessing",
+      "prov:label": "Package the frozen benchmark as public structured tables",
+      "prov:description": "Preserve the anonymous-release benchmark values, replace local absolute paths with relative asset paths, and flatten the frozen JSON into entries (900 rows), pairs (1643 rows), and task3_edit_plans (100 rows) with scripts/make_genscale_parquet.py. Do not impute missing scenario-specific cells. Package 433 image files, retain source references, and provide CSV/Parquet mirrors and SHA-256 checksums. The Croissant update adds documentation without changing benchmark records.",
+      "prov:used": [
+        {
+          "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/3002a005586c105caf5b18c4ddd13db1be458625/GenScale_Benchmark.json"
+        },
+        {
+          "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/3002a005586c105caf5b18c4ddd13db1be458625/scripts/make_genscale_parquet.py"
+        }
+      ],
+      "prov:wasAssociatedWith": [
+        {
+          "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale#genscale-research-team",
+          "@type": "prov:Agent",
+          "prov:label": "GenScale authors / benchmark research team"
+        },
+        {
+          "@id": "https://github.com/lingxiao-li/GenScale",
+          "@type": "prov:SoftwareAgent",
+          "prov:label": "GenScale public construction/serialization scripts"
+        }
+      ]
+    }
+  ]
+}
+```

RAI_REQUIREMENTS.json

@@ -0,0 +1,48 @@
+{
+  "required_fields": [
+    "rai:dataLimitations",
+    "rai:dataBiases",
+    "rai:personalSensitiveInformation",
+    "rai:dataUseCases",
+    "rai:dataSocialImpact",
+    "rai:hasSyntheticData",
+    "prov:wasDerivedFrom",
+    "prov:wasGeneratedBy"
+  ],
+  "all_eight_present": true,
+  "synthetic_boolean_type_correct": true,
+  "metadata_sha256": "8535399233639d590549a6f552a3b36ddc7a42c4a8d9cc953b9c99b48e2a0b77",
+  "data_revision_pinned": "3002a005586c105caf5b18c4ddd13db1be458625",
+  "paper_source": "papers/GenScale_camera_ready_review_v2.pdf, Sec. 3.1-3.3; Appendices A, B and H",
+  "checker_caveat": "The linked checker currently tests six RAI keys; PROV source/activity fields are additionally audited against the conference's eight-topic minimum.",
+  "official_requirements": "https://neurips.cc/Conferences/2026/EvaluationsDatasetsHosting",
+  "official_checker": "https://huggingface.co/spaces/JoaquinVanschoren/croissant-checker",
+  "official_checker_event_id": "531e1a2e5a754fd28ee42c0b1ea557a8",
+  "official_checker_checks": [
+    {
+      "name": "JSON Format Validation",
+      "passed": true,
+      "message": "\u2713\nThe file is valid JSON."
+    },
+    {
+      "name": "Croissant Schema Validation",
+      "passed": true,
+      "message": "\u2713\nThe dataset passes Croissant validation."
+    },
+    {
+      "name": "Responsible AI Metadata",
+      "passed": true,
+      "message": "\u2713\nAll required Responsible AI metadata fields are present."
+    },
+    {
+      "name": "Records Generation Test",
+      "passed": true,
+      "message": "\u2713\nRecord set 'entries' passed validation.\nRecord set 'pairs' passed validation.\nRecord set 'task3_edit_plans' passed validation."
+    }
+  ],
+  "recordset_rows_loaded_locally": {
+    "entries": 900,
+    "pairs": 1643,
+    "task3_edit_plans": 100
+  }
+}

README.md

@@ -65,7 +65,9 @@ The public release contains prompts, object/product metadata, and images. It doe
 
 Machine-assisted construction metadata records Gemini model identifiers where available. The paper's scoring pipeline is provided in the code repository; benchmark rows are not themselves the complete per-model judge-output archive. Trained Rescale checkpoints are distributed separately at [lingxiao2049/Rescale](https://huggingface.co/lingxiao2049/Rescale) under the FLUX non-commercial license. This dataset does not bundle those weights, all generated model outputs, or raw human-study responses. Report bugs through the linked GitHub repository; cite a Hugging Face commit revision for reproducible use.
 
-[`croissant.json`](croissant.json) describes all three structured tables using Croissant with the RAI extension, including collection, missing data, provenance, intended uses, limitations, bias considerations, and personal-information scope.
+[`croissant.json`](croissant.json) is the canonical, author-maintained Croissant metadata file for all three structured tables. It includes all eight minimum RAI topics in the [NeurIPS 2026 hosting guidance](https://neurips.cc/Conferences/2026/EvaluationsDatasetsHosting): limitations, biases, personal/sensitive information, intended uses and construct validity, social impact, a synthetic-data boolean, source datasets, and collection/preprocessing/annotation provenance. Synthetic source prompts and available generator identifiers are documented; missing generation seeds and backend revisions are explicitly identified. CSV sources are pinned to an immutable public dataset revision and have SHA-256 checksums.
+
+The exact file passed all four checks in the [conference-linked Croissant checker](https://huggingface.co/spaces/JoaquinVanschoren/croissant-checker) (JSON, schema, RAI fields, and record generation). See [CROISSANT_VALIDATION_REPORT.txt](CROISSANT_VALIDATION_REPORT.txt) and [RAI_REQUIREMENTS.json](RAI_REQUIREMENTS.json). The checker currently tests six RAI keys; the two required PROV source/activity fields were additionally reviewed against the conference guidance. Local `mlcroissant` validation loaded all 900 / 1,643 / 100 records. **For OpenReview, upload this repository's `croissant.json`, which contains the authored RAI information; an automatically generated metadata export may omit these additions.**
 
 ## License and attribution
 

SHA256SUMS

@@ -1,12 +1,14 @@
 9e75dd981de037ec3769f24f790e126bc5a160b6871f510214e68dc70649aeeb  .gitattributes
 b6f13ae534358591390f90dad18ba39ede2cfef03e5714191041d6248280ebab  CITATION.bib
+af147ddc18742f4e8d98017b22a40161f5ecbc71b5594b4115df53b51bd5ef49  CROISSANT_VALIDATION_REPORT.txt
 ee129f6958f9d82934f8cf62b084a2a316b3ddf9c956e59f9954131ca6bd7477  GenScale_Benchmark.json
 ee129f6958f9d82934f8cf62b084a2a316b3ddf9c956e59f9954131ca6bd7477  GenScale_Benchmark_anonymous_clean.json
 419896aea50c15d6e40c5b4baf4bd346f78223b9a354c7357ad00262afdc08ec  LICENSE
 6e1ed9d527583991c4116281cf30062ec2090af7131e9e4f5c3ece9d0c9fb546  PROCESSING_REPORT.md
-0710117347f2b7e23e1bd8019b54c8fc1dc1be816fa7778c44b3838e1de818ab  README.md
+9fef929c1b059021a6f18a7a5d204be26e1d4668f101748a3b0c596e4cb2fbfe  RAI_REQUIREMENTS.json
+82baf49b79eb073e6abc802c99323dd1cdf7f83ef4f8a0424d132772a5f9544f  README.md
 8b15577d9ecf0404c29969bab5aeafb2dfd9a4fd684ba60ca58d0ab70836ae20  THIRD_PARTY_NOTICES.md
-765db3f4ff45fcb44b382bcbe684fd5a6f1b720b799609e50b5e1bc25bf7cb12  croissant.json
+8535399233639d590549a6f552a3b36ddc7a42c4a8d9cc953b9c99b48e2a0b77  croissant.json
 8cf26356645a165d6bf64438322ab602e1a9d820fc76575e802916f4bae8a7f1  data/genscale_entries.csv
 a34d296a90d556c47939841cfdda0f1b8663483734e82864c68e4f969d508af7  data/genscale_entries.parquet
 0f312e352e9f7cd20bbb6e63f97cd58bab1120f903ada97e2418a35417f26e6d  data/genscale_pairs.csv

VALIDATION.json

@@ -7,7 +7,7 @@
   },
   "image_references": 578,
   "unique_images": 433,
-  "verified_checksums": 451,
+  "verified_checksums": 453,
   "csv_parquet_values_match": true,
   "croissant": {
     "mlcroissant_version": "1.1.1",
@@ -19,7 +19,30 @@
       "task3_edit_plans": 100
     },
     "rai_fields_parsed": true,
-    "validation_errors": 0
+    "validation_errors": 0,
+    "metadata_sha256": "8535399233639d590549a6f552a3b36ddc7a42c4a8d9cc953b9c99b48e2a0b77",
+    "all_recordset_rows_loaded": true,
+    "recordset_definitions_preserved": true,
+    "csv_checksums_preserved": true,
+    "original_benchmark_bytes_preserved": true,
+    "required_rai_fields": [
+      "rai:dataLimitations",
+      "rai:dataBiases",
+      "rai:personalSensitiveInformation",
+      "rai:dataUseCases",
+      "rai:dataSocialImpact",
+      "rai:hasSyntheticData",
+      "prov:wasDerivedFrom",
+      "prov:wasGeneratedBy"
+    ],
+    "all_eight_present": true,
+    "synthetic_boolean_type_correct": true,
+    "provenance_sources": 6,
+    "provenance_activities": 6,
+    "official_checker": "https://huggingface.co/spaces/JoaquinVanschoren/croissant-checker",
+    "official_checker_space_revision": "5a09c4a7740e8e42f982d35217a5310d89d8566c",
+    "official_checker_event_id": "531e1a2e5a754fd28ee42c0b1ea557a8",
+    "official_checker_all_four_passed": true
   },
   "source_repository": "anonymous2049/GenScale",
   "source_revision": "25c38a96151ecda875d03fa3ffa1eae7469528d6",

croissant.json

@@ -44,7 +44,8 @@
     "separator": "cr:separator",
     "source": "cr:source",
     "subField": "cr:subField",
-    "transform": "cr:transform"
+    "transform": "cr:transform",
+    "prov": "http://www.w3.org/ns/prov#"
   },
   "@type": "sc:Dataset",
   "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale",
@@ -92,12 +93,12 @@
     "image editing",
     "benchmark"
   ],
-  "rai:dataCollection": "Constructed from a 100-category common-object size knowledge base, a 1000-product Amazon Berkeley Objects (ABO) candidate pool, selected catalog reference images, API-assisted prompts, and generated editing inputs. Frozen benchmark values are preserved from the anonymous review release.",
+  "rai:dataCollection": "The GenScale research team curated a 100-category size knowledge base from COCO/LVIS category vocabularies and cited physical-size references; 97 non-human categories serve the common-object pool. ABO product dimensions and catalog reference images ground the human-product task. Model-assisted prompts and generated Task 1/2 images provide synthetic content, and 100 erroneous but editable generated scenes become Task 3 inputs. The release contains 433 unique image files (294 Task 2 reference images, 100 Task 3 generated inputs and 39 Task 3 references). It packages existing benchmark records, without new image generation or participant collection. Per-record collection dates/geographic sampling and numeric generation seeds are not retained; the released prompts, source task IDs, product IDs, dimension sources and preserved source images provide the available reconstruction inputs. See paper Sec. 3.1-3.2 and Appendix A.",
   "rai:dataCollectionType": [
-    "Derived physical-size metadata",
-    "Selected catalog images",
-    "Synthetic image inputs",
-    "Machine-assisted prompts"
+    "Curated physical-size references and object vocabularies",
+    "Selected ABO catalog metadata and product reference images",
+    "Machine-assisted generation/editing prompts",
+    "Synthetic generated scenes selected for scale editing"
   ],
   "rai:dataCollectionMissingData": "Scenario-inapplicable fields are null or empty. Only selected benchmark reference/input images are included; original ABO candidate-pool paths require the upstream archive. All nonempty benchmark image/reference paths resolve within this release.",
   "rai:dataCollectionRawData": "ABO: https://amazon-berkeley-objects.s3.amazonaws.com/index.html . Common-object metadata source_reference values are retained. This release includes 433 selected image files and does not redistribute the full original ABO archive.",
@@ -107,31 +108,37 @@
     "No imputation of missing physical-size fields; original numerical annotations retained."
   ],
   "rai:dataAnnotationProtocol": [
-    "Pair records provide intended physical A/B ratios and acceptable intervals; precise editing plans specify target, anchor, direction, and resize factor.",
-    "These tables are benchmark inputs, not the complete human-study or per-model judge-output archive."
+    "Benchmark target labels are physical characteristic-length A/B ratios and acceptable intervals derived from the curated size knowledge base or ABO dimensions with canonical human anchors. The precise-edit table adds the editable object, fixed reference object, resize direction and factor. These are not nine independent human annotations for every released relation.",
+    "The automatic evaluation protocol uses an ordinal 1-5 scale with 3 denoting plausible target scale; the judge uses anchor A and target B with the reciprocal B/A ratio convention. Nine human raters separately calibrated that judge using a custom annotation GUI; consensus uses the mode with median tie breaking. The paper describes the rubric, instructions and agreement analysis in Sec. 3.3 and Appendix B. Those private rater responses are not distributed as benchmark rows.",
+    "Task 3 selection requires a visible and editable scale error and removes missing/unidentifiable objects, duplicate/merged objects, severe artifacts, heavy occlusion/blur and extreme resizing cases. This is quality control for scorable/editable inputs, not an independent fairness or privacy audit."
   ],
   "rai:machineAnnotationTools": [
-    "Gemini API assistance for prompt refinement and construction; where available the exact model identifier is retained in refine_meta.gemini_model.",
-    "The associated code repository provides Gemini v5 scoring scripts and frozen GenScale-Open reference assets."
+    "Gemini API assistance is used for prompt construction/refinement and evaluation; the released refine_meta.gemini_model field records the refinement model where available.",
+    "The preserved Task 3 source provenance identifies FLUX.2 (61 unique inputs) and Qwen-Image-Edit-2511 (39 unique inputs). Source task IDs and prompts are released; exact dated API backends and per-image numeric RNG seeds are not fully retained.",
+    "Human judge-calibration annotations use the custom interface described in paper Appendix B. Public judging and aggregation implementations are in the GenScale code repository."
   ],
   "rai:dataUseCases": [
-    "Research evaluation of relative object scale in image generation and editing.",
-    "Comparison of generated scale relations against supplied physical-size metadata."
+    "Measure physically grounded pairwise relative object scale in image generation and editing, under the characteristic-length definitions and acceptable physical-size intervals in the released metadata.",
+    "Compare common-object generation (S1/S2), human-product generation with metric prompts (S3), and scale correction from selected erroneous generated inputs (S4/S5). The construct is pairwise relative scale, not full 3D geometry, image quality, authenticity, or all aspects of physical realism.",
+    "Evidence for this evaluative use is the paper's physical-size grounding, documented ordinal rubric, and calibration against human consensus (Sec. 3.1-3.3 and Appendix A-B). These establish a scoped evaluation protocol rather than universal measurement validity. Demographic fairness auditing, safety-critical measurement, authentication of images, and general-purpose model training have not been validated by this dataset."
   ],
   "rai:dataLimitations": [
-    "Small English-language benchmark with selected everyday object and product categories.",
-    "Physical metadata can be imperfect and can conflict with prompt prose; generated inputs may contain artifacts.",
-    "Fixed human-body anchors simplify real variation and are not population statistics.",
-    "S4 and S5 share the same 100 source images; the train split label is a storage convention.",
-    "Training checkpoints, complete generated model outputs and raw human-study forms are not included."
+    "900 English-language prompt entries and 1643 pairwise records cover selected common objects and catalog products, not a representative sample of all objects, cultures, bodies or environments.",
+    "Physical metadata and canonical human-anchor lengths are reference assumptions; object sizes vary in the real world. Metadata dimensions can be imperfect and can conflict with prompt prose. Generated inputs can contain visual artifacts. See paper Sec. 3.1 and Appendix H.6.",
+    "S4 and S5 reuse the same 100 source images (61 common-object and 39 human-product cases); their 200 prompt entries are not 200 independent source images. Task 3 selects failed but editable generations and is not a random sample of generator outputs.",
+    "The Hugging Face train split name is a storage convention, not a prescribed model-training split. The 1000-product candidate metadata includes upstream ABO archive paths whose full raw image pool is not bundled. All nonempty benchmark image/reference paths resolve to the 433 packaged images.",
+    "Do not use this benchmark as a safety-critical physical measurement system, an authenticity detector, or evidence of complete 3D/physical correctness. Complete per-model generated outputs and private human annotation responses are outside this dataset package. Rescale weights are released separately at https://huggingface.co/lingxiao2049/Rescale under their own license."
   ],
   "rai:dataBiases": [
-    "Selection reflects the common-object vocabulary and ABO product catalog; demographic and cultural representativeness are not established.",
-    "Model-assisted construction and generated inputs may reflect the biases of the source models. No independent fairness audit is claimed."
+    "Category selection starts from COCO/LVIS vocabularies and favors visually identifiable objects with relatively stable characteristic lengths. The resulting coverage is shaped by those vocabularies and by the authors' curation; ambiguous, highly variable and non-rigid categories are underrepresented.",
+    "Product examples inherit ABO catalog coverage, English listing-title selection, dimensional-data availability and a controlled characteristic-length range. This is not a global or demographic sample.",
+    "Fixed anthropometric anchors simplify variation in age, sex, disability and body size. Prompts and generated human depictions may reflect gender/age descriptors and source-model biases; no representative demographic labels or group-specific validity claims are provided.",
+    "Task 3 deliberately samples visible, scorable, editable scale errors from specified source-model outputs and applies visual quality filtering. Conclusions about correction should account for this selection. The automatic judge can inherit model biases; no independent fairness audit is claimed."
   ],
   "rai:personalSensitiveInformation": [
-    "No rater identities, private annotation forms, contact information or person-identification labels are included.",
-    "Catalog images and generated scenes may depict faces or brands. No independent privacy audit is claimed."
+    "The public fields describe objects, product titles/identifiers, dimensions, prompts and image paths. Images may show human faces/body parts and brands; prompt text can contain gender or age descriptors. These are not verified real-person demographic, identity or biometric labels.",
+    "Private rater identities, contact details, individual annotation forms, health records, political views and religious beliefs are not distributed in this package. No independent privacy audit or guarantee that all unstructured images are free of incidental personal information is claimed.",
+    "The paper describes generic human scale anchors, avoidance of named-person prompts, and annotation tasks that do not request sensitive participant disclosures (Appendix H.3-H.4). This metadata does not infer an IRB/exemption status or annotator demographic distribution."
   ],
   "rai:dataReleaseMaintenancePlan": "Report issues via https://github.com/lingxiao-li/GenScale/issues . Cite a dataset commit revision; checksums are supplied in SHA256SUMS. No fixed update cadence is promised.",
   "distribution": [
@@ -140,7 +147,7 @@
       "@id": "entries_csv",
       "name": "data/genscale_entries.csv",
       "description": "CSV table for GenScale entries. Empty cells denote absent fields.",
-      "contentUrl": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/main/data/genscale_entries.csv",
+      "contentUrl": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/3002a005586c105caf5b18c4ddd13db1be458625/data/genscale_entries.csv",
       "encodingFormat": "text/csv",
       "sha256": "8cf26356645a165d6bf64438322ab602e1a9d820fc76575e802916f4bae8a7f1",
       "contentSize": "635212"
@@ -150,7 +157,7 @@
       "@id": "pairs_csv",
       "name": "data/genscale_pairs.csv",
       "description": "CSV table for GenScale pairs. Empty cells denote absent fields.",
-      "contentUrl": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/main/data/genscale_pairs.csv",
+      "contentUrl": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/3002a005586c105caf5b18c4ddd13db1be458625/data/genscale_pairs.csv",
       "encodingFormat": "text/csv",
       "sha256": "0f312e352e9f7cd20bbb6e63f97cd58bab1120f903ada97e2418a35417f26e6d",
       "contentSize": "1088540"
@@ -160,7 +167,7 @@
       "@id": "task3_edit_plans_csv",
       "name": "data/genscale_task3_edit_plans.csv",
       "description": "CSV table for GenScale task3_edit_plans. Empty cells denote absent fields.",
-      "contentUrl": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/main/data/genscale_task3_edit_plans.csv",
+      "contentUrl": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/3002a005586c105caf5b18c4ddd13db1be458625/data/genscale_task3_edit_plans.csv",
       "encodingFormat": "text/csv",
       "sha256": "0029aa2d231aed3a7d1b28f0cbd75a8a97e08f5498ff5a275f30731a19db5361",
       "contentSize": "47682"
@@ -1511,5 +1518,214 @@
         }
       ]
     }
+  ],
+  "dateModified": "2026-10-17",
+  "rai:hasSyntheticData": true,
+  "rai:dataSocialImpact": "Positive potential: make relative-scale failures in image generation and editing measurable, support reproducible evaluation, and inform more physically plausible product visualization, design tools and simulation. These are potential uses, not independently established deployment outcomes. Negative potential: correcting scale artifacts can make manipulated or synthetic images more convincing and enable misleading product depictions or other deceptive visual communication. The selected English object/product vocabulary and fixed body-size anchors may also encourage overgeneralization to underrepresented people, cultures or physical environments. Mitigations documented in the release include source/task identifiers, explicit synthetic-data disclosure, published limitations, separate asset licenses, and exclusion of private rater records and credentials. The paper recommends research use and disclosure of synthetic editing, rather than treating scale plausibility as proof of authenticity or overall physical correctness. No independent fairness, privacy or misuse audit is claimed. See the paper Appendix H.3 and H.5-H.6.",
+  "prov:wasDerivedFrom": [
+    {
+      "@id": "https://amazon-berkeley-objects.s3.amazonaws.com/index.html",
+      "@type": "prov:Entity",
+      "prov:label": "Amazon Berkeley Objects (ABO)",
+      "prov:description": "Product identifiers, dimensions and selected catalog reference images; original source assets remain subject to ABO attribution and CC BY 4.0."
+    },
+    {
+      "@id": "https://cocodataset.org/",
+      "@type": "prov:Entity",
+      "prov:label": "COCO category vocabulary",
+      "prov:description": "Category names inform the curated object vocabulary. COCO images and complete annotations are not redistributed as this benchmark."
+    },
+    {
+      "@id": "https://www.lvisdataset.org/",
+      "@type": "prov:Entity",
+      "prov:label": "LVIS category vocabulary",
+      "prov:description": "Category names inform the curated object vocabulary. LVIS images and complete annotations are not redistributed as this benchmark."
+    },
+    {
+      "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/3002a005586c105caf5b18c4ddd13db1be458625/metadata/authoritative_kb_3d_100.csv",
+      "@type": "prov:Entity",
+      "prov:label": "GenScale physical-size construction seed table",
+      "prov:description": "100 category-level dimension records, including per-row source_reference citations for anthropometric, biological/food, manufactured-object, traffic and sports-size references."
+    },
+    {
+      "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/3002a005586c105caf5b18c4ddd13db1be458625/metadata/abo_local_sampled_1000_representative_anonymized.csv",
+      "@type": "prov:Entity",
+      "prov:label": "ABO candidate construction seed table",
+      "prov:description": "1000 candidate products and dimensions used in benchmark construction; upstream raw image paths are identifiers for the ABO archive."
+    },
+    {
+      "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/3002a005586c105caf5b18c4ddd13db1be458625/GenScale_Benchmark.json",
+      "@type": "prov:Entity",
+      "prov:label": "Frozen prompt and source-task seed records",
+      "prov:description": "The preserved generation/editing prompts, task identifiers, source task identifiers and physical-ratio metadata reconstruct the available synthetic-data inputs; exact per-image numeric random seeds are not supplied."
+    }
+  ],
+  "prov:wasGeneratedBy": [
+    {
+      "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale#activity-vocabulary-and-sizes",
+      "@type": "prov:Activity",
+      "prov:type": "collection",
+      "prov:label": "Curate object vocabulary and physical-size priors",
+      "prov:description": "Research-team curation of visually identifiable categories with stable physical extent, using cited size references and anthropometric anchors. Source vocabularies and per-row dimension references are preserved; collection timestamps and a geographic sampling frame are not available. Paper Sec. 3.1 and Appendix A.1.",
+      "prov:used": [
+        {
+          "@id": "https://cocodataset.org/"
+        },
+        {
+          "@id": "https://www.lvisdataset.org/"
+        },
+        {
+          "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/3002a005586c105caf5b18c4ddd13db1be458625/metadata/authoritative_kb_3d_100.csv"
+        }
+      ],
+      "prov:wasAssociatedWith": [
+        {
+          "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale#genscale-research-team",
+          "@type": "prov:Agent",
+          "prov:label": "GenScale authors / benchmark research team"
+        }
+      ]
+    },
+    {
+      "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale#activity-product-filtering",
+      "@type": "prov:Activity",
+      "prov:type": "preprocessing",
+      "prov:label": "Filter ABO products and assign human anchors",
+      "prov:description": "Require an English title, valid dimensions and a retrievable main image; convert dimensions to centimeters and use the maximum axis as characteristic length. The paper retains characteristic lengths in [5,250] cm, then assigns hand, face/head, foot/leg or full-body anchors by the documented size bands. Product identifiers and dimensions remain in the released records. Paper Appendix A.3.",
+      "prov:used": [
+        {
+          "@id": "https://amazon-berkeley-objects.s3.amazonaws.com/index.html"
+        },
+        {
+          "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/3002a005586c105caf5b18c4ddd13db1be458625/metadata/abo_local_sampled_1000_representative_anonymized.csv"
+        }
+      ],
+      "prov:wasAssociatedWith": [
+        {
+          "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale#genscale-research-team",
+          "@type": "prov:Agent",
+          "prov:label": "GenScale authors / benchmark research team"
+        }
+      ]
+    },
+    {
+      "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale#activity-prompt-construction",
+      "@type": "prov:Activity",
+      "prov:type": "collection",
+      "prov:label": "Construct machine-assisted benchmark prompts",
+      "prov:description": "Generate common-object prompts and product/human interaction prompts from selected objects, product metadata and dimensions using Gemini-assisted construction/refinement. The frozen prompt text and available refine_meta model identifier are released. Numeric RNG seeds are not documented per prompt. Paper Sec. 3.2 and Appendix A.",
+      "prov:used": [
+        {
+          "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/3002a005586c105caf5b18c4ddd13db1be458625/metadata/authoritative_kb_3d_100.csv"
+        },
+        {
+          "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/3002a005586c105caf5b18c4ddd13db1be458625/metadata/abo_local_sampled_1000_representative_anonymized.csv"
+        },
+        {
+          "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/3002a005586c105caf5b18c4ddd13db1be458625/GenScale_Benchmark.json"
+        }
+      ],
+      "prov:wasAssociatedWith": [
+        {
+          "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale#genscale-research-team",
+          "@type": "prov:Agent",
+          "prov:label": "GenScale authors / benchmark research team"
+        },
+        {
+          "@id": "https://ai.google.dev/gemini-api/docs",
+          "@type": "prov:SoftwareAgent",
+          "prov:label": "Gemini API construction/refinement tools"
+        }
+      ]
+    },
+    {
+      "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale#activity-synthetic-source-images",
+      "@type": "prov:Activity",
+      "prov:type": "collection",
+      "prov:label": "Generate and select Task 3 synthetic image inputs",
+      "prov:description": "The preserved source records identify 61 FLUX.2 common-object images and 39 Qwen-Image-Edit-2511 human-product images. Select failed but scorable/editable pairwise cases, apply the documented object-visibility/artifact/correction-factor quality controls, and create S4/S5 prompts for each of the same 100 source images. Exact API backend revisions and image-level numeric seeds are not fully retained; the images and their source-task prompts/IDs are preserved. Paper Sec. 3.2 and Appendix A.4-A.5.",
+      "prov:used": [
+        {
+          "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/3002a005586c105caf5b18c4ddd13db1be458625/GenScale_Benchmark.json"
+        }
+      ],
+      "prov:wasAssociatedWith": [
+        {
+          "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale#genscale-research-team",
+          "@type": "prov:Agent",
+          "prov:label": "GenScale authors / benchmark research team"
+        },
+        {
+          "@id": "https://bfl.ai/blog/flux-2",
+          "@type": "prov:SoftwareAgent",
+          "prov:label": "FLUX.2 source image generator"
+        },
+        {
+          "@id": "https://huggingface.co/Qwen/Qwen-Image-Edit-2511",
+          "@type": "prov:SoftwareAgent",
+          "prov:label": "Qwen-Image-Edit-2511 source image generator"
+        }
+      ]
+    },
+    {
+      "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale#activity-ratio-and-edit-labels",
+      "@type": "prov:Activity",
+      "prov:type": "annotation",
+      "prov:label": "Derive physical-ratio labels and precise edit plans",
+      "prov:description": "Compute intended physical A/B ratios and acceptable intervals from characteristic sizes. For precise editing, store the target and reference object, rendered ratio, direction and multiplicative resize factor in the released edit-plan records. Separate human judge calibration uses nine raters with the five-point rubric and mode/median consensus described in Sec. 3.3 and Appendix B; it is not nine human labels per benchmark relation. Private annotation forms/rater identities are not part of these tables, and no new demographic or ethics-review claims are made here.",
+      "prov:used": [
+        {
+          "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/3002a005586c105caf5b18c4ddd13db1be458625/metadata/authoritative_kb_3d_100.csv"
+        },
+        {
+          "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/3002a005586c105caf5b18c4ddd13db1be458625/metadata/abo_local_sampled_1000_representative_anonymized.csv"
+        },
+        {
+          "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/3002a005586c105caf5b18c4ddd13db1be458625/GenScale_Benchmark.json"
+        },
+        {
+          "@id": "https://arxiv.org/abs/2609.00525"
+        }
+      ],
+      "prov:wasAssociatedWith": [
+        {
+          "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale#genscale-research-team",
+          "@type": "prov:Agent",
+          "prov:label": "GenScale authors / benchmark research team"
+        },
+        {
+          "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale#calibration-raters",
+          "@type": "prov:Agent",
+          "prov:label": "Nine judge-calibration raters; identities not distributed"
+        }
+      ]
+    },
+    {
+      "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale#activity-public-packaging",
+      "@type": "prov:Activity",
+      "prov:type": "preprocessing",
+      "prov:label": "Package the frozen benchmark as public structured tables",
+      "prov:description": "Preserve the anonymous-release benchmark values, replace local absolute paths with relative asset paths, and flatten the frozen JSON into entries (900 rows), pairs (1643 rows), and task3_edit_plans (100 rows) with scripts/make_genscale_parquet.py. Do not impute missing scenario-specific cells. Package 433 image files, retain source references, and provide CSV/Parquet mirrors and SHA-256 checksums. The Croissant update adds documentation without changing benchmark records.",
+      "prov:used": [
+        {
+          "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/3002a005586c105caf5b18c4ddd13db1be458625/GenScale_Benchmark.json"
+        },
+        {
+          "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale/resolve/3002a005586c105caf5b18c4ddd13db1be458625/scripts/make_genscale_parquet.py"
+        }
+      ],
+      "prov:wasAssociatedWith": [
+        {
+          "@id": "https://huggingface.co/datasets/lingxiao2049/GenScale#genscale-research-team",
+          "@type": "prov:Agent",
+          "prov:label": "GenScale authors / benchmark research team"
+        },
+        {
+          "@id": "https://github.com/lingxiao-li/GenScale",
+          "@type": "prov:SoftwareAgent",
+          "prov:label": "GenScale public construction/serialization scripts"
+        }
+      ]
+    }
   ]
 }