licenses/SOURCE_TERMS.md

# Source-specific terms

This is an academic curation of existing datasets, not a transfer of copyright.

Kvasir-SEG: research and education only; commercial use requires prior written author permission. Cite Jha et al., Kvasir-SEG, MMM 2020. Current original terms: https://datasets.simula.no/kvasir-seg/

PolypGen: preserve both conflicting original notices supplied here. The archive license specifies CC BY-NC-SA 4.0, while its README specifies CC BY 4.0. Follow the stricter archive notice for this academic release; clarify with the original authors for broader use. Cite Ali et al., Scientific Data 10, 75 (2023), https://doi.org/10.1038/s41597-023-01981-y . Our changes are curation, exact duplicate removal, fixed center splits, binary-mask conversion and mask-derived weak task labels. Original image bytes are retained.

CVC-ClinicDB: no CVC image or mask bytes are redistributed in this repository. Only the previously audited external-test identity manifest is included. Original rights remain with Hospital Clinic and CVC. Obtain data from the original author channel and comply with applicable access/use terms: https://pages.cvc.uab.es/CVC-Colon/index.php/databases/ and https://pages.cvc.uab.es/ai4polypnet/wp-content/uploads/2024/05/AgreementData_CVC-1.pdf . The agreement was located but its complete current text could not be retrieved during this publication; this is not a finding that all redistribution is prohibited.



Derived weak text labels and curation contributions are supplied for noncommercial research and education under CC BY-NC-SA 4.0, subject to all original image terms. This does not relicense Kvasir or CVC. https://creativecommons.org/licenses/by-nc-sa/4.0/





## Kvasir-VQA extension



The separate vqa_sft configuration uses original Kvasir-VQA images and atomic annotations by Gautam et al., source revision 26df9125b98cbad664e5b1801aac1f70a02689e2. CC BY-NC 4.0 applies to these source materials. The source card also requests prior permission for competitions/commercial use. Cite https://doi.org/10.48550/arXiv.2409.01437 and https://datasets.simula.no/kvasir-vqa. Fixed English sentence rendering and question paraphrases are our transformations; no x1 LLM-generated labels are included. This does not relicense the original materials under PolypGen terms. Read Kvasir-VQA_SOURCE_CARD.md.