Polling-based Object Probing Evaluation (POPE) on A-OKVQA POPE Random
89.6AccuracyICT w/o image
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ICT w/o imageModel=LLaVA-v1.52024.11 | 89.6 | 89.75 | |
| ICT w/o imageModel=Qwen-VL2024.11 | 89.46 | 89.28 | |
| ICTModel=Qwen-VL2024.11 | 89.46 | 89.03 | |
| ICT w/o objectModel=Qwen-VL2024.11 | 89.26 | 88.95 | |
| VCDModel=Qwen-VL2024.11 | 89.22 | 89.01 | |
| ICTModel=LLaVA-v1.52024.11 | 89.2 | 89.41 | |
| ICT w/o objectModel=LLaVA-v1.52024.11 | 89 | 88.91 | |
| OPERAModel=Qwen-VL2024.11 | 88.19 | 88.43 | |
| OPERAModel=LLaVA-v1.52024.11 | 88.02 | 84.59 | |
| RegularModel=Qwen-VL2024.11 | 86.67 | 85.59 | |
| VCDModel=LLaVA-v1.52024.11 | 86.15 | 86.34 | |
| RegularModel=LLaVA-v1.52024.11 | 83.45 | 82.56 |