Multimodal Emotion Reasoning on MEP-Bench
80.17RecallHumanOmni-V2‡
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| HumanOmni-V2‡Training Strategy=Oracle reference2026.06 | 80.17 | 48.4 | 37.6 | |
| OPPOTraining Strategy=Ours2026.06 | 70.44 | 76.4 | 50.6 | |
| BaselineTraining Strategy=Ours2026.06 | 58 | 59.2 | 33 | |
| AffectGPT-R1Training Strategy=RL Baseline2026.06 | 54.85 | 54.2 | 30 | |
| Qwen2.5-OmniTraining Strategy=SFT Baseline2026.06 | 53 | 57 | 33.8 | |
| AffectGPTTraining Strategy=SFT Baseline2026.06 | 49.14 | 35 | 26 |