CVPR - How to identify if an accepted paper has ethical issues (plagiarism)? [D]

I recently found a paper accepted to CVPR 2026 reproduced many technical details from my paper submitted to arXiV on June 2025 (5 months before the CVPR 2026 submission deadline).

Apart from technical similarities (they rephrased / reframed the term / key ideas), the CVPR paper uses exactly same equation without changes to any notations from our paper without proper citation. Several figures show high similarities in style and pipeline.

We tried to contact authors from the CVPR paper, but they framed the technical similarity as "general method" so no need to cite. While they admitted that they refer to our paper for figure design, writing style, and equation, they can only update the arXiv version of their paper (the CVPR camera ready deadline has passed), claiming that they are "inspired" by us. Basically they would not do anything to their proceeding paper.

I am wondering how CVPR identify the plagiarism between their accepted papers and arXiv papers? Will it be considered as plagiarism only if they reproduce a published work?

Thanks for any advice!

Attached part of the reproduction:

Our arXiv work applied a multi-turn extension on the basic GRPO algorithm (with notation changes). The CVPR paper directly adopted the exact same equation without citation.

Our arXiv paper

The CVPR paper