Object Hallucination Mitigation on Object-Hal
58CHAIRs ScoreLLaVA-RLHF
Evaluation results
| Method | Links | ||
|---|---|---|---|
| LLaVA-RLHFBackbone=LLaVA-v1.5-7B, Datasize=122k2026.05 | 58 | 15.61 | |
| LLaVA-v1.5-7BBackbone=LLaVA-v1.5-7B2026.05 | 55.67 | 15.96 | |
| LLaVA-v1.5-13BBackbone=LLaVA-v1.5-13B2026.05 | 51 | 13.71 | |
| POVIDBackbone=LLaVA-v1.5-7B, Datasize=17k2026.05 | 50.67 | 15.28 | |
| HALVABackbone=LLaVA-v1.5-13B, Datasize=22.5k2026.05 | 45.4 | 12.8 | |
| LLaVA-RLHFBackbone=LLaVA-v1.5-13B, Datasize=122k2026.05 | 44.67 | 11.83 | |
| Qwen2.5-VL-3BBackbone=Qwen2.5-VL-3B2026.05 | 43.2 | 12.2 | |
| HALVABackbone=LLaVA-v1.5-7B, Datasize=21.5k2026.05 | 41.4 | 11.7 | |
| mDPOBackbone=LLaVA-v1.5-7B, Datasize=10k2026.05 | 35.7 | 9.8 | |
| DPOBackbone=Qwen2.5-VL-3B, Datasize=5.7k2026.05 | 30.6 | 10.7 | |
| UE-DPOBackbone=Qwen2.5-VL-3B, Datasize=5.7k2026.05 | 16.7 | 5.6 | |
| RLAIF-VBackbone=LLaVA-v1.5-7B, Datasize=16k2026.05 | 16 | 3.7 | |
| UE-DPOBackbone=LLaVA-v1.5-7B, Datasize=5.7k, Training Data Source=RLHF-V Data2026.05 | 13.72 | 6.69 | |
| UE-DPOBackbone=LLaVA-v1.5-13B, Datasize=5.7k2026.05 | 13.7 | 6.57 | |
| UE-DPOBackbone=LLaVA-v1.5-7B, Datasize=16k, Training Data Source=RLAIF-V Data2026.05 | 11.62 | 5.16 |