arXiv preprint / cs.CV

Beyond Global Editing: Per-Instance Disentangled Subspaces for Training-Free Hallucination Mitigation in LVLMs

作者：Ali Cheraghian, Hamidreza Dastmalchi, Hamed Barzamini, Morteza Saberi, Mojtaba Golzan, Shafin Rahman, Hossein Rahmani · 会议/期刊：arXiv preprint · 方向：cs.CV · 发布日期：2026-08-11

2026-08-11视觉语言
新标签打开 AlphaXiv
Beyond Global Editing: Per-Instance Disentangled Subspaces for Training-Free Hallucination Mitigation in LVLMs figure
AlphaXiv 中文概览（可滚动查看）