Vision–Language Models (VLMs) have advanced video action recognition, yet most existing approaches still rely on text as a static semantic cue. In real video data, however, the primary challenge often lies in the noise (dominant backgrounds, scene biases, and temporally irrelevant motions) that obscure discriminative p...
Jiang-Huai Li, Ling-Xiao Yang, Jian-Huang Lai et al.· Proceedings of the 34th ACM...· 0 citations
COSMO replaces expert-to-expert guidance with co-adaptation through an anchored shared consensus and achieves state-of-the-art performance under matched VLM backbones, indicating that it better balances the retention of valid source-derived evidence with the absorption of complementary VLM evidence.
Bo Li, Junjie Peng, Xiaohua Xie et al.· 0 citations
GRKV (Global Regression for KV Cache), a training-free KV-cache merging method that directly minimizes the discrepancy between compressed-cache and full-cache attention outputs, is proposed.
Junjie Peng, You Wu, Haoyi Wu et al.· arXiv.org· 2 citations
A dual-stream image-to-video architecture built on a pretrained diffusion-transformer video generator that improves VideoPhy-2 Physical-Commonsense and Video-Quality scores over the frozen backbone by up to 8.75 and 4.65 points, outperforms a leading open competitor, and is preferred by human raters in a blind study.
Rui Su, Yuanzhi Liang, Xiaohua Xie et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.