1 paper
Shaoxuan Li, Zhixuan Zhao, Hanze Deng +9
We introduce PerceptionComp, a manually annotated benchmark for complex, long-horizon, perception-centric video reasoning. PerceptionComp is designed so that no single moment is su…