Object Hallucination on POPE Popular
93.33AccuracyORCA
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| ORCABase LVLM=MiniGPT-4 (13B), Agent Backbone=LLaMA-3.2-Vision-Instruct (11B), Visual Backbone=EVA2025.09 | 93.33 | — | — | 93.01 | — | — | |
| ORCABase LVLM=Qwen-VL-Chat (13B), Agent Backbone=LLaMA-3.2-Vision-Instruct (11B), Visual Backbone=OpenCLIP2025.09 | 92.67 | — | — | 92.31 | — | — | |
| ORCABase LVLM=mPlug-Owl (7B), Agent Backbone=LLaMA-3.2-Vision-Instruct (11B), Visual Backbone=CLIP2025.09 | 92 | — | — | 91.84 | — | — | |
| LogicCheckGPTModel=LLaVA-1.52024.02 | 91.67 | — | — | 91.4 | — | — | |
| REVIS2026.02 | 90.3 | — | 86.4 | 89.91 | — | — | |
| LogicCheckGPTModel=QWEN-VL-Chat2024.02 | 89.67 | — | — | 88.64 | — | — | |
| REVISModel=LLaVA-NeXT, max_new_token=5122026.02 | 89.57 | — | 84.07 | 88.96 | — | — | |
| CASTBackbone=LLaVA-NeXT2026.05 | 89.53 | — | — | 89.24 | — | — | |
| VTIModel=LLaVA-NeXT, max_new_token=5122026.02 | 88.9 | — | 81.6 | 88.03 | — | — | |
| SelfCheckModel=LLaVA-1.52024.02 | 88.67 | — | — | 88.59 | — | — | |
| RegularModel=LLaVA-NeXT, Decoding=Standard greedy, max_new_token=5122026.02 | 88.67 | — | 81.2 | 87.75 | — | — | |
| M3IDModel=LLaVA-NeXT, max_new_token=5122026.02 | 88.67 | — | 81.2 | 87.75 | — | — | |
| AGLAModel=LLaVA-NeXT, max_new_token=5122026.02 | 88.67 | — | 81.2 | 87.75 | — | — | |
| ONLYModel=LLaVA-NeXT, max_new_token=5122026.02 | 88.6 | — | 81.93 | 87.79 | — | — | |
| Only2026.02 | 88.37 | — | 80.2 | 87.33 | — | — | |
| CASTBackbone=LLaVA-1.5-7b2026.05 | 88.32 | — | — | 87.95 | — | — | |
| VCDModel=LLaVA-NeXT, max_new_token=5122026.02 | 88.3 | — | 82.13 | 87.53 | — | — | |
| SpecEyesBase Model=Thyme, Aggregation Strategy=mean2026.03 | 88.3 | — | — | — | — | 1.8 | |
| SpecEyesBase Model=Thyme, Aggregation Strategy=bottom2026.03 | 88.3 | — | — | — | — | 1.81 | |
| SpecEyesBase Model=Thyme, Aggregation Strategy=min2026.03 | 88.3 | — | — | — | — | 1.78 | |
| VCD2026.02 | 88.27 | — | 80.93 | 87.34 | — | — | |
| Regulardecoding=standard greedy2026.02 | 88.23 | — | 79.73 | 87.14 | — | — | |
| M3ID2026.02 | 88.23 | — | 79.73 | 87.14 | — | — | |
| UNIFIED-IO 2Model Size=XXL2023.12 | 88.17 | 89.13 | 86.93 | 88.02 | 48.77 | — | |
| VCD + CNSModel Backbone=Qwen-VL2026.01 | 88.17 | — | — | 86.78 | — | — | |
| MGAPBackbone=LLaVA v1.5 7B2026.05 | 88.1 | 91.5 | — | 87.59 | — | — | |
| VGAModel=Qwen2.5-VL2025.11 | 88.06 | — | — | 87.81 | — | — | |
| AGLA2026.02 | 88.03 | — | 79.33 | 86.89 | — | — | |
| MiniGPT-4 + ORCALVLM=MiniGPT-4, Method=+ ORCA, Attack Scenario=Adversarial Illusions2025.09 | 88 | — | — | 86.67 | — | — | |
| LLaVA1.5 w/ HACLzero-shot=true, HACL=true2023.12 | 87.94 | 97.23 | 79.31 | 87.36 | 45.03 | — | |
| PAIModel=Qwen2.5-VL2025.11 | 87.93 | — | — | 87.65 | — | — | |
| VAFModel=Qwen2.5-VL2025.11 | 87.9 | — | — | 87.41 | — | — | |
| VanillaModel=Qwen2.5-VL2025.11 | 87.88 | — | — | 87.24 | — | — | |
| TARACModel=Qwen2.5-VL2025.11 | 87.88 | — | — | 87.22 | — | — | |
| Osprey-7B*Model Size=7B2023.12 | 87.83 | 89.94 | 85.2 | 87.5 | 47.37 | — | |
| SpecEyesBase Model=Thyme, Aggregation Strategy=log2026.03 | 87.8 | — | — | — | — | 1.67 | |
| CASTBackbone=Qwen-VL-Chat2026.05 | 87.73 | — | — | 86.84 | — | — | |
| PADELVLM Model=Qwen-VL2026.02 | 87.72 | — | — | 86.12 | — | — | |
| DeCoBackbone=LLaVA v1.5 7B2026.05 | 87.72 | 89.36 | — | 87.31 | — | — | |
| PRISM2025.02 | 87.7 | — | — | — | — | — | |
| SelfCheckModel=QWEN-VL-Chat2024.02 | 87.67 | — | — | 87.87 | — | — | |
| MiniGPT-4 + ORCALVLM=MiniGPT-4, Variant=+ ORCA2025.09 | 87.67 | — | — | 86.35 | — | — | |
| mPlug-Owl + ORCALVLM=mPlug-Owl, Method=+ ORCA, Attack Scenario=Adversarial Illusions2025.09 | 87.67 | — | — | 86.64 | — | — | |
| VTI2026.02 | 87.57 | — | 80.4 | 86.61 | — | — | |
| ICONS2025.02 | 87.5 | — | — | — | — | — | |
| ICDBackbone=LLaVA v1.5 7B2026.05 | 87.5 | 89.04 | — | 87.24 | — | — | |
| UNIFIED-IO 2Model Size=XL2023.12 | 87.47 | 95.69 | 78.47 | 86.23 | 41 | — | |
| LLaVA1.5zero-shot=true2023.12 | 87.46 | 95.17 | 78.93 | 86.29 | 41.46 | — | |
| VTIBackbone=LLaVA-1.5-7b2026.05 | 87.36 | — | — | 86.69 | — | — | |
| LUREModel=QWEN-VL-Chat2024.02 | 87.33 | — | — | 87.16 | — | — | |
| VTIBackbone=LLaVA-NeXT2026.05 | 87.33 | — | — | 87.16 | — | — | |
| HyperLLaVALLM=Vicuna-7B, Activated parameters=7B, Zero-shot=true2024.03 | 87.3 | — | — | 86.2 | 42.4 | — | |
| V-STARDecoding=Sampling2026.04 | 87.3 | — | — | 86.6 | — | — | |
| RGPT2024.03 | 87.2 | 95.44 | 78.13 | 85.92 | 40.93 | — | |
| LLaVA-1.5LLM=Vicuna-7B, Activated parameters=7B, Zero-shot=true2024.03 | 87.2 | — | — | 86.1 | 41.9 | — | |
| OPERABackbone=LLaVA-NeXT2026.05 | 87.16 | — | — | 87.68 | — | — | |
| REVISModel=LLaVA-1.5, max_new_token=5122026.02 | 87.13 | — | 83.33 | 86.63 | — | — | |
| MoDBackbone=LLaVA v1.5 7B2026.05 | 87.03 | 87.96 | — | 86.91 | — | — | |
| VCDBackbone=LLaVA-NeXT2026.05 | 87.01 | — | — | 87.7 | — | — | |
| LUREModel=LLaVA-1.52024.02 | 87 | — | — | 87.05 | — | — | |
| SpecEyesBase Model=DeepEyes, Aggregation Strategy=mean2026.03 | 87 | — | — | — | — | 2.1 | |
| SpecEyesBase Model=DeepEyes, Aggregation Strategy=bottom2026.03 | 87 | — | — | — | — | 2.08 | |
| SpecEyesBase Model=DeepEyes, Aggregation Strategy=min2026.03 | 87 | — | — | — | — | 2.15 | |
| OursDecoding Strategy=Greedy, Base Model=Qwen-VL2025.01 | 86.92 | — | — | 86.55 | — | — | |
| VARLVLM Model=Qwen-VL2026.02 | 86.89 | — | — | 84.92 | — | — | |
| RC-DPOModel=MM-Eureka2026.05 | 86.8 | — | — | 85.69 | — | — | |
| OPERABackbone=LLaVA-1.52026.06 | 86.8 | — | — | — | — | — | |
| LION2023.11 | 86.77 | 91.69 | 80.87 | 85.94 | — | — | |
| Vanilla + CNSModel Backbone=Qwen-VL2026.01 | 86.77 | — | — | 84.85 | — | — | |
| SpecEyesBase Model=DeepEyes, Aggregation Strategy=log2026.03 | 86.7 | — | — | — | — | 1.95 | |
| V-STARDecoding=Greedy2026.04 | 86.7 | — | — | 86 | — | — | |
| FoxBackbone=LLaVA-1.52026.06 | 86.7 | — | — | — | — | — | |
| OursDecoding Strategy=Sampling, Base Model=Qwen-VL2025.01 | 86.67 | — | — | 86.34 | — | — | |
| PAIBackbone=LLaVA-NeXT2026.05 | 86.65 | — | — | 86.99 | — | — | |
| OPERABackbone=LLaVA-1.5-7b2026.05 | 86.64 | — | — | 86.62 | — | — | |
| CausalMMBackbone=LLaVA-1.52026.06 | 86.63 | — | — | — | — | — | |
| VGAModel=LLaVA-Next2025.11 | 86.62 | — | — | 86.24 | — | — | |
| VAFModel=LLaVA-Next2025.11 | 86.58 | — | — | 85.95 | — | — | |
| OursBackbone=PixelReasoner-7B2026.03 | 86.5 | — | — | 85.3 | — | — | |
| VCDModel=MM-Eureka2026.05 | 86.43 | — | — | 84.91 | — | — | |
| MGAPBackbone=Qwen3-VL 8B2026.05 | 86.4 | 84.13 | — | 86.84 | — | — | |
| vanillaModel=QWEN-VL-Chat2024.02 | 86.33 | — | — | 86.37 | — | — | |
| mPlug-Owl + ORCALVLM=mPlug-Owl, Variant=+ ORCA2025.09 | 86.33 | — | — | 84.98 | — | — | |
| OursDecoding Strategy=Greedy, Base Model=LLaVA-1.52025.01 | 86.33 | — | — | 86 | — | — | |
| VanillaBackbone=LLaVA v1.5 7B2026.05 | 86.23 | 83.28 | — | 86.8 | — | — | |
| mPLUG-Owl22023.12 | 86.2 | 89.46 | 82.06 | 85.6 | 45.86 | — | |
| PTIBase LVLM=Qwen-VL-Chat2026.04 | 86.17 | — | — | 84.89 | — | — | |
| VISTABase LVLM=DeepSeek-VL-Chat2026.04 | 86.13 | — | — | 84.58 | — | — | |
| OPA-DPOModel=MM-Eureka2026.05 | 86.13 | — | — | 84.49 | — | — | |
| VAFLVLM Model=Qwen-VL2026.02 | 86.12 | — | — | 84.28 | — | — | |
| Full-Finetune2025.02 | 86.1 | — | — | — | — | — | |
| COINCIDE2025.02 | 86.1 | — | — | — | — | — | |
| DeCoBackbone=Qwen3-VL 8B2026.05 | 86.01 | 83.02 | — | 83.81 | — | — | |
| LocVLM-B2024.04 | 86 | 79.7 | 93.9 | 86.3 | 58.9 | — | |
| Qwen-VL-ChatParameters=13B, Backbone=OpenCLIP, Setup=Standalone2025.09 | 86 | — | — | 86 | — | — | |
| OursDecoding Strategy=Sampling, Base Model=LLaVA-1.52025.01 | 86 | — | — | 85.7 | — | — | |
| SpecReasonBase Model=Thyme2026.03 | 85.97 | — | — | — | — | 0.43 | |
| PAIModel=LLaVA-Next2025.11 | 85.95 | — | — | 84.7 | — | — | |
| HalTrapperBackbone=Qwen3-VL 8B2026.05 | 85.94 | 82.2 | — | 86.06 | — | — | |
| SIDModel=MM-Eureka2026.05 | 85.93 | — | — | 84.35 | — | — |