Object Hallucination on POPE Adversarial
90AccuracyLogicCheckGPT
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| LogicCheckGPTModel=LLaVA-1.52024.02 | 90 | — | — | 89.58 | — | — | — | — | — | |
| LogicCheckGPTModel=QWEN-VL-Chat2024.02 | 89 | — | — | 88 | — | — | — | — | — | |
| SelfCheckModel=LLaVA-1.52024.02 | 88.67 | — | — | 88.27 | — | — | — | — | — | |
| SCRBackbone=InternVL2-7B2026.02 | 88.5 | — | — | — | — | — | 11.2 | 4.6 | — | |
| SCRBackbone=Qwen2-VL-7B2026.02 | 88.2 | — | — | — | — | — | 11.4 | 5.6 | — | |
| CRoPS†Backbone=InternVL2-7B2026.02 | 88 | — | — | — | — | — | 10.8 | 5 | — | |
| CogVLM-ChatLLM=Vicuna-7B2023.11 | 87.9 | — | — | — | — | — | — | — | — | |
| REVIS2026.02 | 87.8 | — | 86.4 | 87.63 | — | — | — | — | — | |
| Unified-IO2LLM=UIO-2XXL2023.11 | 87.7 | — | — | — | — | — | — | — | — | |
| SelfCheckModel=QWEN-VL-Chat2024.02 | 87.67 | — | — | 87.54 | — | — | — | — | — | |
| OA-VCDBackbone=InternVL2-7B2026.02 | 87.6 | — | — | — | — | — | 12 | 3.8 | — | |
| V-STARDecoding=Sampling2026.04 | 87.6 | — | — | 87 | — | — | — | — | — | |
| SIDBackbone=InternVL2-7B2026.02 | 87.4 | — | — | — | — | — | 12.2 | 3.6 | — | |
| SPHINX-2kLLM=LLaMA2 13B2023.11 | 87.2 | — | — | — | — | — | — | — | — | |
| VCDBackbone=InternVL2-7B2026.02 | 87.2 | — | — | — | — | — | 12.4 | 3.4 | — | |
| REVISModel=LLaVA-NeXT, max_new_token=5122026.02 | 87.13 | — | 84.07 | 86.73 | — | — | — | — | — | |
| Regulardecoding=standard greedy2026.02 | 87.1 | — | 79.87 | 86.09 | — | — | — | — | — | |
| M3ID2026.02 | 87.1 | — | 79.87 | 86.09 | — | — | — | — | — | |
| CRoPS†Backbone=Qwen2-VL-7B2026.02 | 87.1 | — | — | — | — | — | 12.6 | 4.4 | — | |
| VCD2026.02 | 87.03 | — | 80.93 | 86.19 | — | — | — | — | — | |
| LUREModel=QWEN-VL-Chat2024.02 | 87 | — | — | 87.62 | — | — | — | — | — | |
| Only2026.02 | 86.93 | — | 80.13 | 85.98 | — | — | — | — | — | |
| V-STARDecoding=Greedy2026.04 | 86.9 | — | — | 86.3 | — | — | — | — | — | |
| RegularModel=LLaVA-NeXT, Decoding=Standard greedy, max_new_token=5122026.02 | 86.87 | — | 81.2 | 86.08 | — | — | — | — | — | |
| M3IDModel=LLaVA-NeXT, max_new_token=5122026.02 | 86.87 | — | 81.2 | 86.08 | — | — | — | — | — | |
| AGLAModel=LLaVA-NeXT, max_new_token=5122026.02 | 86.87 | — | 81.2 | 86.08 | — | — | — | — | — | |
| SCRBackbone=Qwen-VL2026.02 | 86.8 | — | — | — | — | — | 12.9 | 5.6 | — | |
| OA-VCDBackbone=Qwen2-VL-7B2026.02 | 86.8 | — | — | — | — | — | 12.9 | 4.1 | — | |
| AGLA2026.02 | 86.73 | — | 79.27 | 85.67 | — | — | — | — | — | |
| VLI†Backbone=InternVL2-7B2026.02 | 86.7 | — | — | — | — | — | 12.9 | 2.9 | — | |
| vanillaModel=QWEN-VL-Chat2024.02 | 86.67 | — | — | 86.67 | — | — | — | — | — | |
| VTIModel=LLaVA-NeXT, max_new_token=5122026.02 | 86.57 | — | 81.6 | 85.86 | — | — | — | — | — | |
| LLaVA1.5 w/ HACLzero-shot=true, HACL=true2023.12 | 86.54 | 93.01 | 79.52 | 85.73 | 46.33 | — | — | — | — | |
| SCRBackbone=LLaVA-1.5-7B2026.02 | 86.5 | — | — | — | — | — | 13.1 | 5.7 | — | |
| VCDBackbone=Qwen2-VL-7B2026.02 | 86.5 | — | — | — | — | — | 13.1 | 3.9 | — | |
| Uniform-SmoothBackbone=InternVL2-7B2026.02 | 86.5 | — | — | — | — | — | 13.2 | 2.6 | — | |
| mPLUG-Owl2LLM=LLaMA2-7B2023.11 | 86.2 | — | — | — | — | — | — | — | — | |
| SCRBackbone=Cham-7B2026.02 | 86.2 | — | — | — | — | — | 13.5 | 6 | — | |
| SIDBackbone=Qwen2-VL-7B2026.02 | 86.2 | — | — | — | — | — | 13.5 | 3.5 | — | |
| VCDModel=LLaVA-NeXT, max_new_token=5122026.02 | 86.13 | — | 82.07 | 85.55 | — | — | — | — | — | |
| ONLYModel=LLaVA-NeXT, max_new_token=5122026.02 | 86.13 | — | 81.87 | 85.52 | — | — | — | — | — | |
| VCD + CNSModel Backbone=Qwen-VL2026.01 | 86.12 | — | — | 84.28 | — | — | — | — | — | |
| RC-DPOModel=MM-Eureka2026.05 | 86.07 | — | — | 85.02 | — | — | — | — | — | |
| VLI†Backbone=Qwen2-VL-7B2026.02 | 86 | — | — | — | — | — | 13.7 | 3.3 | — | |
| OPERABackbone=InternVL2-7B2026.02 | 86 | — | — | — | — | — | 13.5 | 2.3 | — | |
| CASTBackbone=LLaVA-NeXT2026.05 | 85.97 | — | — | 86.07 | — | — | — | — | — | |
| LLaVA1.5zero-shot=true2023.12 | 85.93 | 91.78 | 78.93 | 84.87 | 43 | — | — | — | — | |
| LLaVA-1.5LLM=Vicuna-7B2023.11 | 85.9 | — | — | — | — | — | — | — | — | |
| LLaVA-1.5LLM=Vicuna-13B2023.11 | 85.9 | — | — | — | — | — | — | — | — | |
| OA-VCDBackbone=Qwen-VL2026.02 | 85.9 | — | — | — | — | — | 13.8 | 4.7 | — | |
| Vanilla + CNSModel Backbone=Qwen-VL2026.01 | 85.86 | — | — | 83.92 | — | — | — | — | — | |
| CRoPS†Backbone=Qwen-VL2026.02 | 85.8 | — | — | — | — | — | 13.9 | 4.6 | — | |
| UNIFIED-IO 2Model Size=XL2023.12 | 85.77 | 92.01 | 78.33 | 84.62 | 42.57 | — | — | — | — | |
| DoLaBackbone=InternVL2-7B2026.02 | 85.7 | — | — | — | — | — | 13.8 | 2 | — | |
| RGPT2024.03 | 85.67 | 91.99 | 78.13 | 84.5 | 42.47 | — | — | — | — | |
| VCDModel=MM-Eureka2026.05 | 85.67 | — | — | 84.21 | — | — | — | — | — | |
| HyperLLaVALLM=Vicuna-7B, Activated parameters=7B, Zero-shot=true2024.03 | 85.6 | — | — | 84.7 | 44.1 | — | — | — | — | |
| SIDBackbone=Qwen-VL2026.02 | 85.6 | — | — | — | — | — | 14.1 | 4.4 | — | |
| Uniform-SmoothBackbone=Qwen2-VL-7B2026.02 | 85.6 | — | — | — | — | — | 14.2 | 2.8 | — | |
| SIDModel=MM-Eureka2026.05 | 85.4 | — | — | 83.85 | — | — | — | — | — | |
| LION2023.11 | 85.37 | 88.69 | 81.07 | 84.71 | — | — | — | — | — | |
| VTIBackbone=LLaVA-NeXT2026.05 | 85.35 | — | — | 84.52 | — | — | — | — | — | |
| Osprey-7B*Model Size=7B2023.12 | 85.33 | 85.43 | 85.2 | 85.31 | 49.87 | — | — | — | — | |
| LUREModel=LLaVA-1.52024.02 | 85.33 | — | — | 86.25 | — | — | — | — | — | |
| CRoPS†Backbone=LLaVA-1.5-7B2026.02 | 85.3 | — | — | — | — | — | 14.4 | 4.4 | — | |
| OPA-DPOModel=MM-Eureka2026.05 | 85.3 | — | — | 83.71 | — | — | — | — | — | |
| VTI2026.02 | 85.27 | — | 80.4 | 84.51 | — | — | — | — | — | |
| CRoPS†Backbone=Cham-7B2026.02 | 85.2 | — | — | — | — | — | 14.5 | 5 | — | |
| OA-VCDBackbone=LLaVA-1.5-7B2026.02 | 85.2 | — | — | — | — | — | 14.5 | 4.3 | — | |
| VCDBackbone=Qwen-VL2026.02 | 85.2 | — | — | — | — | — | 14.4 | 4.1 | — | |
| OPERABackbone=LLaVA-NeXT2026.05 | 85.2 | — | — | 85.54 | — | — | — | — | — | |
| PADELVLM Model=Qwen-VL2026.02 | 85.12 | — | — | 83.08 | — | — | — | — | — | |
| LLaVA-1.5LLM=Vicuna-7B, Activated parameters=7B, Zero-shot=true2024.03 | 85.1 | — | — | 84.2 | 44 | — | — | — | — | |
| OPERABackbone=Qwen2-VL-7B2026.02 | 85.1 | — | — | — | — | — | 14.6 | 2.4 | — | |
| VanillaModel=MM-Eureka2026.05 | 85.03 | — | — | 83.23 | — | — | — | — | — | |
| OA-VCDBackbone=Cham-7B2026.02 | 85 | — | — | — | — | — | 14.7 | 4.8 | — | |
| VLI†Backbone=Qwen-VL2026.02 | 84.9 | — | — | — | — | — | 14.8 | 3.7 | — | |
| VCDBackbone=LLaVA-NeXT2026.05 | 84.8 | — | — | 85.23 | — | — | — | — | — | |
| RC-DPOModel=R1-Onevision2026.05 | 84.8 | — | — | 83.24 | — | — | — | — | — | |
| VCDBackbone=Cham-7B2026.02 | 84.6 | — | — | — | — | — | 15.1 | 4.4 | — | |
| VCDBackbone=LLaVA-1.5-7B2026.02 | 84.6 | — | — | — | — | — | 15 | 3.8 | — | |
| Uniform-SmoothBackbone=Qwen-VL2026.02 | 84.6 | — | — | — | — | — | 15.1 | 3.4 | — | |
| MGAPBackbone=LLaVA v1.5 7B2026.05 | 84.59 | 85.06 | — | 84.46 | — | — | — | — | — | |
| VLI†Backbone=LLaVA-1.5-7B2026.02 | 84.5 | — | — | — | — | — | 15.2 | 3.6 | — | |
| SIDBackbone=LLaVA-1.5-7B2026.02 | 84.5 | — | — | — | — | — | 15.2 | 3.6 | — | |
| DoLaBackbone=Qwen2-VL-7B2026.02 | 84.5 | — | — | — | — | — | 15.2 | 1.8 | — | |
| RLAIF-VModel=MM-Eureka2026.05 | 84.5 | — | — | 82.57 | — | — | — | — | — | |
| SIDBackbone=Cham-7B2026.02 | 84.4 | — | — | — | — | — | 15.3 | 4.2 | — | |
| CASTBackbone=Qwen-VL-Chat2026.05 | 84.33 | — | — | 83.92 | — | — | — | — | — | |
| PAIBackbone=LLaVA-NeXT2026.05 | 84.32 | — | — | 83.68 | — | — | — | — | — | |
| LLaVA-1.5†LLM=L-7B, zero-shot=true, image_encoder=LanguageBind-Image2023.11 | 84.3 | — | — | 83.2 | 43.5 | — | — | — | — | |
| OursBackbone=PixelReasoner-7B2026.03 | 84.3 | — | — | 83.1 | — | — | — | — | — | |
| CASTBackbone=LLaVA-1.5-7b2026.05 | 84.27 | — | — | 84.41 | — | — | — | — | — | |
| UNIFIED-IO 2Model Size=XXL2023.12 | 84.17 | 82.17 | 87.27 | 84.64 | 53.1 | — | — | — | — | |
| mPLUG-Owl22023.12 | 84.12 | 85.54 | 82.13 | 83.8 | 48 | — | — | — | — | |
| VL-CogitoDecoding=Greedy2026.04 | 84.1 | — | — | 83.5 | — | — | — | — | — | |
| OPERABackbone=Qwen-VL2026.02 | 84 | — | — | — | — | — | 15.6 | 2.9 | — | |
| VARLVLM Model=Qwen-VL2026.02 | 83.96 | — | — | 81.82 | — | — | — | — | — | |
| VLI†Backbone=Cham-7B2026.02 | 83.8 | — | — | — | — | — | 15.8 | 3.7 | — | |
| VanillaBackbone=InternVL2-7B2026.02 | 83.8 | — | — | — | — | — | 15.8 | — | — |