The academic peer review process has a blind spot, and this CVPR 2026 case exposes it clearly. When a paper reuses equations verbatim without citation, that is not "inspiration" or a "general method." It is plagiarism, regardless of whether the original work appeared in a peer-reviewed venue or sat on arXiv. The community needs to stop pretending that the absence of formal publication somehow makes copying acceptable.
The situation described here is not ambiguous. The CVPR paper adopted exact equations with unchanged notations, borrowed figure designs, and mirrored pipeline structures. The authors admitted referencing the arXiv work for style and figures, yet they drew the line at citation. That distinction is convenient, but it collapses under scrutiny. You cannot claim stylistic debt while denying conceptual borrowing, especially when the mathematics is identical. The technical detail is the contribution. If the equations are the same, the idea is the same, and credit is owed.
What matters most for researchers in this position is the enforcement gap. The CVPR authors offered to update only their arXiv version because the camera ready deadline passed. That response treats the proceeding paper as immutable while ignoring that the damage was already done. The original authors are left with a published paper that appears to lack originality, while the copying authors get to claim plausible deniability through post-hoc edits. This is not how rigorous science should operate. Conferences need a formal mechanism for post-acceptance ethical reviews, not just a desk rejection process that runs before the camera ready deadline.
For anyone who has been on the submitting side of this exchange, the practical takeaway is uncomfortable but clear. Your work on arXiv is a vulnerability, not a shield. You must document your submission timestamps, keep detailed version histories, and be prepared to escalate to the conference ethics chairs with evidence. Do not accept "we were inspired" as a final answer. Push for a formal retraction or correction of the proceedings version if the evidence supports you. The system is not designed to protect you, so you have to build your own case. And for the community, the demand is straightforward: if an equation is identical, the citation must follow, or the paper should not stand.