Video reasoning requires understanding how events unfold over time, yet videos with accurate temporal annotations are expensive to collect. Self-play offers an alternative by letting a model generate and solve its own tasks without manual annotation. However, existing video self-play methods train on fixed sets of reco...
Jing-Min Zhu, Yan-Yan Zou, Xiuyu Li et al.· 1 citation
RSE-Bench is introduced, to the authors' knowledge the first benchmark for evaluating skill self-evolution under ordinary, non-adversarial noise, and a method-agnostic framework supporting consistent noise injection and process diagnostics of actual exposure and skill updates is introduced.
Tao Yu, Zhong-Tian Luo, Yu-Chen Long et al.· 0 citations
OPIS is introduced, an input-grounded benchmark that strictly anchors the assessment to a fixed set of object instances from the initial observation for evaluating multi-object memory in video world models and demonstrates that preserving the particular object instances in the input is considerably harder than generati...
Hao Wang, Tao Yu, Liu-Zhou Zhang et al.· arXiv.org· 1 citation
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.