KVSmooth： Mitigating Hallucination in Multi-modal Large Language Models through Key-Value Smoothing
(CVPR 2026)
More Tokens are Better but Not Necessary in Forensic Vision-Language Models
(arxiv 2026) 为多媒体取证任务中的MLLMs设计的免训练视觉 token 压缩框架。 深大、哈工深、中山大学网安
Mitigating the Object Hallucination of LVLM via Adaptive Factual-Guided Activation Editing
（ICLR 2026） 又是加强版VTI。