CVPR - How to identify if an accepted paper has ethical issues (plagiarism)? [D]
Accepted paper copies equations and figures without citation...
A researcher has publicly detailed a case of potential plagiarism involving a paper accepted to CVPR 2026, one of the top computer vision conferences. The accuser's paper, submitted to arXiv in June 2025, introduced a multi-turn extension of the GRPO algorithm with specific notation changes. The CVPR paper, submitted five months later, reproduced the exact same equations without citation, and its figures and pipeline showed high stylistic similarity.
When confronted, the CVPR authors admitted to being 'inspired' by the arXiv work, referencing it for figure design and writing style, but refused to update the conference proceedings because the camera-ready deadline had passed. They offered only to update their arXiv version. This case highlights a critical gap in plagiarism enforcement: many conferences only review against published proceedings, not earlier preprints, leaving researchers vulnerable to uncredited reuse of their work.
- The CVPR 2026 paper copied exact equations from a June 2025 arXiv preprint without citation.
- Figures and pipeline style showed high similarity, admitted by the authors as 'inspiration'.
- Authors refused to update the proceedings, citing the camera-ready deadline, despite admitting similarities.
Why It Matters
This case challenges how top conferences handle plagiarism from preprints, affecting research integrity and credit.