D3-Omni (OmniBias / OmniJudge paper)
Diagnostics for multimodal LLM-as-judge bias across image, video, and speech. The benchmark in the paper is named D3-Omni, not OmniBias.
Failure modes D3-Omni (OmniBias / OmniJudge paper) catches
Position bias
Verbosity bias
Judge miscalibration
Starter command
A starting point, not a guarantee. Commands change between releases, so confirm against the official README.
# paper: arxiv.org/abs/2608.24160
# check the paper for code and data availabilitySources
- 1. D3-Omni (OmniBias / OmniJudge paper) official page or repository · Guangzheng Hu and coauthors · Paper · Sources checked