Object-level Hallucination Evaluation on Obj-Hal
54.4CHAIRs ScoreLLaVA-v1.5-7B
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| LLaVA-v1.5-7BBackbone=LLaVA-v1.5-7B2026.06 | 54.4 | 15.8 | |
| VCDBackbone=LLaVA-v1.5-13B2026.06 | 53.6 | 15.3 | |
| VCDBackbone=LLaVA-v1.5-7B2026.06 | 51.2 | 15 | |
| OPERABackbone=LLaVA-v1.5-7B2026.06 | 50.8 | 14.3 | |
| LLaVA-v1.5-13BBackbone=LLaVA-v1.5-13B2026.06 | 49.8 | 14.6 | |
| DPOBackbone=LLaVA-v1.5-7B2026.06 | 49 | 13 | |
| DPOBackbone=LLaVA-v1.5-13B2026.06 | 43.9 | 12.4 | |
| Qwen2.5-VL-3BBackbone=Qwen2.5-VL-3B2026.06 | 43.2 | 12.2 | |
| OPERABackbone=LLaVA-v1.5-13B2026.06 | 42.6 | 13.2 | |
| VCDBackbone=Qwen2.5-VL-3B2026.06 | 41.4 | 12.3 | |
| SIMABackbone=LLaVA-v1.5-7B2026.06 | 40.9 | 10.4 | |
| mDPOBackbone=LLaVA-v1.5-7B2026.06 | 35.7 | 9.8 | |
| ADAPT-TFBackbone=LLaVA-v1.5-7B2026.06 | 35.6 | 11.8 | |
| ADAPT-TFBackbone=LLaVA-v1.5-13B2026.06 | 35.4 | 11.6 | |
| VCDBackbone=Qwen2.5-VL-7B2026.06 | 35.1 | 10 | |
| ADAPTBackbone=LLaVA-v1.5-13B2026.06 | 34.8 | 10.6 | |
| Qwen2.5-VL-7BBackbone=Qwen2.5-VL-7B2026.06 | 32.3 | 9.6 | |
| ADAPTBackbone=LLaVA-v1.5-7B2026.06 | 30.7 | 8.5 | |
| DPOBackbone=Qwen2.5-VL-3B2026.06 | 30.6 | 10.7 | |
| DPOBackbone=Qwen2.5-VL-7B2026.06 | 29.6 | 11.7 | |
| SPINBackbone=Qwen2.5-VL-7B2026.06 | 28.6 | 7 | |
| ADAPT-TFBackbone=Qwen2.5-VL-7B2026.06 | 27.8 | 8.2 | |
| ADAPT-TFBackbone=Qwen2.5-VL-3B2026.06 | 19.8 | 8.1 | |
| ADAPTBackbone=Qwen2.5-VL-7B2026.06 | 16.8 | 5.6 | |
| ADAPTBackbone=Qwen2.5-VL-3B2026.06 | 15.2 | 5.3 |