Multimodal Hallucination Evaluation on R-Bench
66.68DisIC-VCO
Evaluation results
| Method | Links | ||
|---|---|---|---|
| IC-VCOBackbone=LLaVA-OneVision-Qwen2-7B, Contrastive Sample Source=Contrastive Sample Editing2026.05 | 66.68 | 72.08 | |
| LLaVA-OneVision-Qwen2-7BBackbone=LLaVA-OneVision-Qwen2-7B2026.05 | 66.16 | 71.96 | |
| IC-VCOBackbone=LLaVA-OneVision-Qwen2-7B, Contrastive Sample Source=Synthetic2026.05 | 65.25 | 71.05 | |
| V-DPOBackbone=LLaVA-OneVision-Qwen2-7B, Contrastive Sample Source=Contrastive Sample Editing2026.05 | 64.7 | 68.88 | |
| mDPOBackbone=LLaVA-OneVision-Qwen2-7B, Contrastive Sample Source=Contrastive Sample Editing2026.05 | 64.5 | 70.1 | |
| S-VCOBackbone=LLaVA-OneVision-Qwen2-7B, Contrastive Sample Source=Contrastive Sample Editing2026.05 | 64.3 | 69.69 | |
| SymMPOBackbone=LLaVA-OneVision-Qwen2-7B, Contrastive Sample Source=Synthetic2026.05 | 64.24 | 70.45 | |
| DPOBackbone=LLaVA-OneVision-Qwen2-7B, Contrastive Sample Source=Contrastive Sample Editing2026.05 | 64.05 | 68.43 | |
| SymMPOBackbone=LLaVA-OneVision-Qwen2-7B, Contrastive Sample Source=Contrastive Sample Editing2026.05 | 63.89 | 70.3 | |
| V-DPOBackbone=LLaVA-OneVision-Qwen2-7B, Contrastive Sample Source=Synthetic2026.05 | 63.64 | 69.44 | |
| S-VCOBackbone=LLaVA-OneVision-Qwen2-7B, Contrastive Sample Source=Synthetic2026.05 | 63.44 | 69.64 | |
| DPOBackbone=LLaVA-OneVision-Qwen2-7B, Contrastive Sample Source=Synthetic2026.05 | 63.23 | 69.03 | |
| mDPOBackbone=LLaVA-OneVision-Qwen2-7B, Contrastive Sample Source=Synthetic2026.05 | 63.03 | 70.25 |