Vision-Language Models (VLMs) can generate rich video captions, yet often misidentify which person performs an action or which limb is involved, particularly across camera cuts. Improving these details requires evaluation and training that distinguish missing information from incorrect assertions. We introduce FlexBenc...
Ya-Nan Wang, Ting-Song Li, Kai-Xun Jiang et al.· 0 citations
Recent multimodal deep search agents have advanced tool use and visual content retrieval, yet they still struggle to convert intermediate visual evidence into actionable guidance for subsequent search, limiting their effectiveness in long-horizon settings. To address this gap, we present Neos, a multimodal deep search...
Ekko He, Jack Hong, Shi-Lin Yan et al.· 0 citations
Existing VLA pruning strategies primarily select individual visual tokens according to task-level semantic relevance, while overlooking the spatial information required for robotic manipulation. To examine this limitation, we construct a simple Stride baseline that uniformly samples tokens along the flattened one-dimen...
Jia-Yu Chen, Shuyong Gao, Jing-Kai Jia et al.· 0 citations
Theoretically, the OPD framework is lifted from discrete tokens to continuous-state Markov processes, deriving a closed-form per-step KL objective that unifies both stochastic SDE and deterministic ODE refinement via mean-matching, which provides lower variance and better generality compared to conventional PPO-style p...