Learning to Prefer Reliably: Error-Augmented Emotion Preference Optimization with Calibrated Fusion
Experiments on the MER2026-EmoPrefer Challenge dataset and the error-augmented dataset demonstrate that EAPO improves emotion preference prediction and enhances the robustness of MLLM judges when evaluating fluent descriptions that conflict with the video's multimodal emotional evidence.
Zilong Huang, Junyi Peng, Junjie Li et al.
· 0 citations