ResOT operates in two stages.
Training-free method repairs object hallucination in LVLMs instead of suppressing it
ResOT aligns hallucinated token representations with faithful distributions via Gaussian optimal transport in a residual subspace, cutting hallucination while preserving caption quality.
Research Lab
Chen Zhao · Xingping Dong · Jiachun Shi · Liang Peng · Chong Wang · Zhen Lei · +2 more
Wuhan University · Institute of Automation, Chinese Academy of Sciences
Research Digest··2 min read
The authors propose ResOT, a training-free, inference-time method that mitigates object hallucination in large vision-language models by repairing, rather than suppressing, hallucination-related hidden representations.
Why this paper
From Institute of Automation, Chinese Academy of Sciences and Wuhan University
In one line
ResOT repairs hallucinated representations by aligning them to faithful distributions in a residual subspace, reducing object hallucination without suppressing useful information.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§