Object Hallucination Probing on GQA POPE Popular
86.07AccuracyMESA
Evaluation results
| Method | Links | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| MESAModel=Qwen-VL2026.04 | 86.07 | — | — | — | — | — | — | — | — | — | 92.73 | 78.27 | 84.89 | |
| VAFModel=Qwen-VL2026.04 | 85.43 | — | — | — | — | — | — | — | — | — | 92.61 | 72.53 | 81.35 | |
| ICTModel=Qwen-VL2026.04 | 85.31 | — | — | — | — | — | — | — | — | — | 91.59 | 77.86 | 84.15 | |
| SchroMindModel=LLaVA-1.5-7B2026.02 | 84.83 | — | — | — | — | — | — | — | — | — | — | — | 84.93 | |
| ICT w/o objectModel=LLaVA-v1.52024.11 | 84.7 | — | — | — | — | — | — | — | — | — | — | — | 84.45 | |
| ICTModel=LLaVA-v1.52024.11 | 84.7 | — | — | — | — | — | — | — | — | — | — | — | 84.78 | |
| ScalpelBase Model=LLaVA-1.5-7B2026.02 | 84.57 | — | — | — | — | — | — | — | — | — | — | — | 85.31 | |
| ICT w/o imageModel=LLaVA-v1.52024.11 | 84.23 | — | — | — | — | — | — | — | — | — | — | — | 83.92 | |
| ICT w/o objectModel=Qwen-VL2024.11 | 83.93 | — | — | — | — | — | — | — | — | — | — | — | 82.95 | |
| VTIModel=Qwen-VL2026.04 | 83.67 | — | — | — | — | — | — | — | — | — | 95.01 | 71.07 | 81.31 | |
| MESAModel=LLaVA-v1.52026.04 | 82.93 | — | — | — | — | — | — | — | — | — | 79.37 | 89 | 83.91 | |
| ICTModel=LLaVA-v1.52026.04 | 82.86 | — | — | — | — | — | — | — | — | — | 82.51 | 84.93 | 83.7 | |
| ICTModel=Qwen-VL2024.11 | 82.63 | — | — | — | — | — | — | — | — | — | — | — | 82.22 | |
| VCDModel=Qwen-VL2024.11 | 82.23 | — | — | — | — | — | — | — | — | — | — | — | 82.11 | |
| OPERAModel=Qwen-VL2024.11 | 81.97 | — | — | — | — | — | — | — | — | — | — | — | 82.12 | |
| VCDModel=Qwen-VL2026.04 | 81.83 | — | — | — | — | — | — | — | — | — | 80.45 | 84.09 | 82.23 | |
| VTIModel=LLaVA-v1.52026.04 | 81.73 | — | — | — | — | — | — | — | — | — | 79.53 | 85.47 | 82.39 | |
| ICT w/o imageModel=Qwen-VL2024.11 | 81.43 | — | — | — | — | — | — | — | — | — | — | — | 80.86 | |
| VAFModel=LLaVA-v1.52026.04 | 81.16 | — | — | — | — | — | — | — | — | — | 78.52 | 85.8 | 82 | |
| ICTBase Model=LLaVA-1.5-7B2026.02 | 80.86 | — | — | — | — | — | — | — | — | — | — | — | 82.64 | |
| ICTModel=LLaVA-1.5-7B2026.02 | 80.86 | — | — | — | — | — | — | — | — | — | — | — | 82.64 | |
| ICDModel=Qwen-VL2026.04 | 80.8 | — | — | — | — | — | — | — | — | — | 91.77 | 67.66 | 77.89 | |
| VCDModel=LLaVA-v1.52024.11 | 80.73 | — | — | — | — | — | — | — | — | — | — | — | 82.24 | |
| VCDBase Model=LLaVA-1.5-7B2026.02 | 80.73 | — | — | — | — | — | — | — | — | — | — | — | 82.24 | |
| VCDModel=LLaVA-1.5-7B2026.02 | 80.73 | — | — | — | — | — | — | — | — | — | — | — | 82.24 | |
| VCDModel=LLaVA-v1.52026.04 | 80.73 | — | — | — | — | — | — | — | — | — | 76.26 | 89.24 | 82.24 | |
| OPERAModel=LLaVA-v1.52024.11 | 79.27 | — | — | — | — | — | — | — | — | — | — | — | 81.83 | |
| OPERABase Model=LLaVA-1.5-7B2026.02 | 79.27 | — | — | — | — | — | — | — | — | — | — | — | 82.11 | |
| OPERAModel=LLaVA-1.5-7B2026.02 | 79.27 | — | — | — | — | — | — | — | — | — | — | — | 82.11 | |
| Instruction Contrastive Decoding (ICD)Backbone=InstructBLIP2024.03 | 78.8 | — | — | — | — | — | — | — | — | — | 75.15 | 87.53 | 80.87 | |
| Visual Contrastive Decoding (VCD)Backbone=InstructBLIP2024.03 | 78.57 | — | — | — | — | — | — | — | — | — | 74.62 | 86.61 | 80.17 | |
| ICDModel=LLaVA-v1.52026.04 | 78.56 | — | — | — | — | — | — | — | — | — | 77.95 | 79.66 | 78.79 | |
| RegularModel=LLaVA-v1.52024.11 | 78.17 | — | — | — | — | — | — | — | — | — | — | — | 80.73 | |
| VanillaBase Model=LLaVA-1.5-7B2026.02 | 78.17 | — | — | — | — | — | — | — | — | — | — | — | 78.37 | |
| RegularModel=LLaVA-1.5-7B2026.02 | 78.17 | — | — | — | — | — | — | — | — | — | — | — | 78.37 | |
| NulluModel=LLaVA-v1.52026.04 | 77.63 | — | — | — | — | — | — | — | — | — | 72.08 | 90.2 | 80.13 | |
| RegularModel=Qwen-VL2024.11 | 75.99 | — | — | — | — | — | — | — | — | — | — | — | 74.84 | |
| AVISCModel=LLaVA-1.5-7B2026.02 | 74.8 | — | — | — | — | — | — | — | — | — | — | — | 79.17 | |
| VanillaModel=Qwen-VL2026.04 | 74.76 | — | — | — | — | — | — | — | — | — | 98.18 | 50.46 | 66.66 | |
| NulluModel=Qwen-VL2026.04 | 74.71 | — | — | — | — | — | — | — | — | — | 98.17 | 50.33 | 66.54 | |
| Standard DecodingBackbone=InstructBLIP2024.03 | 73.87 | — | — | — | — | — | — | — | — | — | 60.63 | 84.69 | 76.42 | |
| VanillaModel=LLaVA-v1.52026.04 | 73.3 | — | — | — | — | — | — | — | — | — | 66.4 | 94.33 | 77.93 | |
| M3IDModel=LLaVA-1.5-7B2026.02 | 72.83 | — | — | — | — | — | — | — | — | — | — | — | 77.58 | |
| Instruction Contrastive Decoding (ICD)Backbone=miniGPT42024.03 | 62.84 | — | — | — | — | — | — | — | — | — | 61.09 | 80.54 | 69.48 | |
| Visual Contrastive Decoding (VCD)Backbone=miniGPT42024.03 | 62.14 | — | — | — | — | — | — | — | — | — | 61.14 | 72.26 | 66.24 | |
| Regular DecodingModel=Qwen-VL, Decoding=Regular2026.01 | 58.87 | — | — | — | — | — | — | — | — | — | 59.87 | 53.8 | 56.67 | |
| VCDModel=Qwen-VL, Decoding=VCD2026.01 | 58.83 | — | — | — | — | — | — | — | — | — | 58.65 | 56.73 | 57.68 | |
| SDCDModel=Qwen-VL, Decoding=SDCD2026.01 | 57.47 | — | — | — | — | — | — | — | — | — | 57.22 | 59.2 | 58.19 | |
| Standard DecodingBackbone=miniGPT42024.03 | 57.19 | — | — | — | — | — | — | — | — | — | 58.55 | 60.81 | 59.66 | |
| BaselineBackbone=LLaVA-1.52025.06 | — | 83.9 | 69.6 | 64.03 | 64.53 | 71.53 | 63.4 | — | — | 66.62 | — | — | — | |
| BaselineBackbone=Qwen-VL-Chat2025.06 | — | 85.8 | 81.9 | 72.17 | 75.07 | — | — | 59.2 | 70.4 | 71.93 | — | — | — | |
| CLAIMBackbone=LLaVA-1.52025.06 | — | 81.57 | 78.5 | 77.93 | 79.6 | 80.57 | — | — | — | 79.63 | — | — | — | |
| CLAIMBackbone=Qwen-VL-Chat2025.06 | — | 84.43 | 80.6 | 82.6 | 61.3 | — | — | 78.07 | — | 77.4 | — | — | — | |
| VCDBackbone=LLaVA-1.52025.06 | — | — | 73.6 | 67.5 | 65.23 | 71.87 | 66.2 | — | — | 68.88 | — | — | — | |
| VCDBackbone=Qwen-VL-Chat2025.06 | — | — | 82.2 | 73.73 | 74.17 | — | — | 54.7 | 76.87 | 72.33 | — | — | — |