Object Hallucination Probing on OKVQA POPE Popular
85AccuracyScalpel
Evaluation Results
| Method | Links | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| ScalpelBase Model=LLaVA-1.5-7B2026.02 | 85 | — | — | — | — | — | — | — | — | — | — | — | 85.18 | |
| ICTBase Model=LLaVA-1.5-7B2026.02 | 83.4 | — | — | — | — | — | — | — | — | — | — | — | 84.45 | |
| OPERABase Model=LLaVA-1.5-7B2026.02 | 83.22 | — | — | — | — | — | — | — | — | — | — | — | 84.67 | |
| VCDBase Model=LLaVA-1.5-7B2026.02 | 81.85 | — | — | — | — | — | — | — | — | — | — | — | 82.82 | |
| Instruction Contrastive Decoding (ICD)Backbone=InstructBLIP2024.03 | 81.64 | — | — | — | — | — | — | — | — | — | 78.5 | 88.77 | 83.32 | |
| VanillaBase Model=LLaVA-1.5-7B2026.02 | 79.9 | — | — | — | — | — | — | — | — | — | — | — | 79.59 | |
| Visual Contrastive Decoding (VCD)Backbone=InstructBLIP2024.03 | 79.78 | — | — | — | — | — | — | — | — | — | 76 | 87.05 | 81.15 | |
| Standard DecodingBackbone=InstructBLIP2024.03 | 76.19 | — | — | — | — | — | — | — | — | — | 72.16 | 85.28 | 78.17 | |
| Instruction Contrastive Decoding (ICD)Backbone=miniGPT42024.03 | 62.81 | — | — | — | — | — | — | — | — | — | 61.62 | 75.78 | 67.97 | |
| Visual Contrastive Decoding (VCD)Backbone=miniGPT42024.03 | 62.22 | — | — | — | — | — | — | — | — | — | 62.23 | 68.55 | 65.24 | |
| Standard DecodingBackbone=miniGPT42024.03 | 60.75 | — | — | — | — | — | — | — | — | — | 60.67 | 68.84 | 64.5 | |
| BaselineBackbone=LLaVA-1.52025.06 | — | 86.97 | 74.5 | 63.27 | 69.4 | 77.03 | 67.73 | — | — | 70.39 | — | — | — | |
| BaselineBackbone=Qwen-VL-Chat2025.06 | — | 88.7 | 85.6 | 67.9 | 75.43 | — | — | 58.87 | 75.27 | 72.61 | — | — | — | |
| CLAIMBackbone=LLaVA-1.52025.06 | — | 85 | 83.4 | 81.87 | 84.27 | 77.97 | — | — | — | 82.5 | — | — | — | |
| CLAIMBackbone=Qwen-VL-Chat2025.06 | — | 88.1 | 85.7 | 83.9 | 68.5 | — | — | 86.03 | — | 82.45 | — | — | — | |
| VCDBackbone=LLaVA-1.52025.06 | — | — | 77.57 | 67.1 | 68.4 | 77.03 | 68.2 | — | — | 71.66 | — | — | — | |
| VCDBackbone=Qwen-VL-Chat2025.06 | — | — | 84.17 | 71.57 | 73.3 | — | — | 52.4 | 75.63 | 71.41 | — | — | — |