Visual Hallucination Evaluation on POPE MS-COCO Adversarial sampling (val)
85.48AccuracyRVCD
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| RVCDModel=mPLUG-Owl2, Setting=Adversarial2025.05 | 85.48 | 84.46 | 86.97 | 85.7 | — | |
| RVCDModel=LLaVA-1.5, Setting=Adversarial2025.05 | 85.36 | 84.17 | 87.09 | 85.61 | — | |
| ICT w/o objectModel=Qwen-VL2024.11 | 85.23 | — | — | 84.62 | — | |
| ICT w/o imageModel=Qwen-VL2024.11 | 85.13 | — | — | 84.35 | — | |
| ICTModel=Qwen-VL2024.11 | 84.96 | — | — | 84.42 | — | |
| OPERAModel=Qwen-VL2024.11 | 84.78 | — | — | 83.45 | — | |
| ICTModel=LLaVA-v1.52024.11 | 84.43 | — | — | 83.74 | — | |
| VCDModel=Qwen-VL2024.11 | 84.26 | — | — | 83.9 | — | |
| ICT w/o imageModel=LLaVA-v1.52024.11 | 84.13 | — | — | 83.31 | — | |
| ICT w/o objectModel=LLaVA-v1.52024.11 | 83.63 | — | — | 82.13 | — | |
| RVCDModel=MiniGPT-4, Setting=Adversarial2025.05 | 83.06 | 82.92 | 83.37 | 83.13 | — | |
| RegularModel=Qwen-VL2024.11 | 82.26 | — | — | 80.37 | — | |
| OPERAModel=LLaVA-v1.52024.11 | 81.24 | — | — | 81.38 | — | |
| VCDModel=LLaVA-v1.52024.11 | 80.88 | — | — | 81.33 | — | |
| RegularModel=LLaVA-v1.52024.11 | 78.96 | — | — | 77.57 | — | |
| Beam SearchModel=mPLUG-Owl2, Setting=Adversarial2025.05 | 74.28 | 67.56 | 93.42 | 78.41 | — | |
| Beam SearchModel=LLaVA-1.5, Setting=Adversarial2025.05 | 73.23 | 66.5 | 93.64 | 77.77 | — | |
| OPERAModel=mPLUG-Owl2, Setting=Adversarial2025.05 | 72.75 | 65.9 | 94.34 | 77.59 | — | |
| DREAMLLM-7Bzero-shot=true, backbone=7B2023.09 | 72.63 | 67.07 | 88.93 | 76.47 | 66.3 | |
| InstructBLIP-14Bzero-shot=true, backbone=14B2023.09 | 72.1 | 65.13 | 95.13 | 77.32 | 73.03 | |
| OPERAModel=LLaVA-1.5, Setting=Adversarial2025.05 | 69.98 | 63.26 | 95.36 | 76.06 | — | |
| VCDModel=mPLUG-Owl2, Setting=Adversarial2025.05 | 69.65 | 63.46 | 92.66 | 75.33 | — | |
| DoLaModel=MiniGPT-4, Setting=Adversarial2025.05 | 69 | 69.76 | 67.1 | 68.39 | — | |
| HALCModel=MiniGPT-4, Setting=Adversarial2025.05 | 69 | 69.76 | 67.1 | 68.39 | — | |
| DoLaModel=mPLUG-Owl2, Setting=Adversarial2025.05 | 68.42 | 61.78 | 96.72 | 75.39 | — | |
| HALCModel=mPLUG-Owl2, Setting=Adversarial2025.05 | 68.4 | 61.77 | 96.72 | 75.38 | — | |
| GreedyModel=mPLUG-Owl2, Setting=Adversarial2025.05 | 68.2 | 61.6 | 96.8 | 75.28 | — | |
| VCDModel=LLaVA-1.5, Setting=Adversarial2025.05 | 66.31 | 60.42 | 94.56 | 73.73 | — | |
| DoLaModel=LLaVA-1.5, Setting=Adversarial2025.05 | 66.2 | 60.05 | 96.81 | 74.12 | — | |
| HALCModel=LLaVA-1.5, Setting=Adversarial2025.05 | 66.2 | 60.05 | 96.81 | 74.12 | — | |
| GreedyModel=LLaVA-1.5, Setting=Adversarial2025.05 | 65.92 | 59.84 | 96.82 | 73.97 | — | |
| MiniGPT-4-14Bzero-shot=true, backbone=14B2023.09 | 65.17 | 61.19 | 82.93 | 70.42 | 67.77 | |
| OPERAModel=MiniGPT-4, Setting=Adversarial2025.05 | 61.95 | 57.57 | 91.02 | 70.53 | — | |
| Beam SearchModel=MiniGPT-4, Setting=Adversarial2025.05 | 61.7 | 57.45 | 90.4 | 70.25 | — | |
| VCDModel=MiniGPT-4, Setting=Adversarial2025.05 | 59.32 | 56.73 | 78.61 | 65.9 | — | |
| GreedyModel=MiniGPT-4, Setting=Adversarial2025.05 | 56.72 | 53.82 | 94.68 | 68.63 | — | |
| mPLUG-Owl-7Bzero-shot=true, backbone=7B2023.09 | 50.67 | 50.34 | 99.33 | 66.82 | 98.67 | |
| MMGPT-7Bzero-shot=true, backbone=7B2023.09 | 50 | 50 | 100 | 66.67 | 100 | |
| LLaVA-13Bzero-shot=true, backbone=13B2023.09 | 49.7 | 49.85 | 99.07 | 66.32 | 99.37 |