shivank21/mmconflict-editable-values-1k

MMConflict Editable Values 2K

This dataset contains 2,000 source images with visible atomic values for multimodal conflict research. It has 100 images in each of 20 categories. Every image comes from a photograph, scan, captured website, software screenshot, or page of a source document. The dataset does not contain generated images or project-rendered examples.

Each row records the source, source URL, license, attribution, visible value, question, and a candidate box around the value.

Categories and conflict uses

The conflict examples below describe suitable benchmark cases. The source images are seeds. They are not prebuilt conflict pairs.

Conflict taxonomy

These images can support the five MMConflict causes.

• —C1 covers a value that changed over time.
• —C2 covers sources that report different values.
• —C3 covers values that belong to different variants or interpretations.
• —C4 covers a conflict created or hidden by OCR or chart-reading errors.
• —C5 covers missing, irrelevant, or incomplete evidence.

Annotation note

CORD item prices and totals come from structured source annotations. Commons rows keep the file-level license, creator, and source page. DocLayNet and XFUND provide document annotations. For other categories, OCR proposes the visible value and candidate box. A person should confirm every OCR proposal before using it in a final benchmark or image-editing pipeline.

Licensing

This is a mixed-license collection. The source_license, source_url, and attribution fields apply to each row. Users must follow the license on every image. Do not treat the full repository as CC0 or as commercially reusable.

The collection includes Public Domain, CC0, CC BY, CC BY-SA, CC BY-NC-SA 4.0, CDLA-Permissive-1.0, and other compatible licenses recorded per row. The Klarna and XFUND subsets allow noncommercial use and require ShareAlike. Research use should remain noncommercial unless every selected row has a license that permits the intended commercial use.

For Wikimedia Commons, the collector reads the license and creator from each file through the MediaWiki image information API. It accepts only Public Domain, CC0, CC BY, and CC BY-SA files. It rejects unknown, NoDerivatives, NonCommercial, custom, and GFDL-only files. Commons does not provide one license that applies to every file.

The scientific-chart category uses only the real-image part of Scientific Chart QA 17K. It joins each image to the source license manifest and accepts only CC0, CC BY, or CC BY-SA figures. Synthetic chart rows and figures with unclear, NonCommercial, or NoDerivatives licenses are excluded.

Open Food Facts product-image pixels use CC BY-SA 3.0. This covers nutrition, front-package, and ingredient-panel images. Ingredient-detection annotations use CC BY-SA 4.0. The front-image-classification dataset declares AGPL-3.0 for its classification compilation. Product photos can contain packaging graphics with separate rights, so users should keep source attribution and ShareAlike terms.

Images are stored as RGB JPEGs, so source files are re-encoded and some categories are cropped or resized. Row-level provenance notes identify these changes. Medication images are for document research and are not medical advice.

Yelp images are excluded because the Yelp Open Dataset terms do not permit this public redistribution. Rico, ScreenQA, CharXiv, Open Prices proof photos, and other sources with unclear underlying image rights are also excluded.

Fields

• —image_id
• —file_name
• —category
• —type
• —source_dataset
• —source_id
• —source_url
• —source_license
• —attribution
• —is_synthetic, always false
• —entity
• —item_name
• —value_type
• —visible_value
• —normalized_value
• —unit
• —currency
• —editable_bbox, in [x, y, width, height] pixels
• —question
• —provenance_notes

Source links

• —CORD v2
• —Open Food Facts ingredient-detection
• —Wikimedia Commons reuse rules
• —Klarna Product Page Dataset
• —Drug Name Detection Dataset
• —DocLayNet-small
• —Scientific Chart QA 17K
• —XFUND