Zero-Shot Evaluation on Image Task Suite (VQAv2, GQA, TextVQA, VizWiz, ScienceQA, POPE, OK-VQA)
62.4Accuracy (%)MMER
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| MMEREvaluation Protocol=Merged2025.05 | 62.4 | 100.3 | |
| Original MLLMsEvaluation Protocol=Zero-shot2025.05 | 62.23 | — | |
| PCB-MergingEvaluation Protocol=Merged2025.05 | 56.02 | 90 | |
| TA (w/ DARE)Evaluation Protocol=Merged2025.05 | 55.47 | 88.4 | |
| TAEvaluation Protocol=Merged2025.05 | 54.89 | 87.5 | |
| TIES (w/ DARE)Evaluation Protocol=Merged2025.05 | 54.73 | 87.2 | |
| TIESEvaluation Protocol=Merged2025.05 | 54.1 | 86.2 | |
| NaiveMC (w/ DARE)Evaluation Protocol=Merged2025.05 | 53.61 | 85.3 | |
| NaiveMCEvaluation Protocol=Merged2025.05 | 52.56 | 83.6 |