Document Question Answering on InfoVQA (test)
72.64AccuracyOracle
Evaluation results
| Method | Links | |
|---|---|---|
| OracleModel=GPT-4o2026.02 | 72.64 | |
| OracleModel=MiniCPM-V 2.62026.02 | 68.8 | |
| MARAModel=GPT-4o2026.02 | 68.02 | |
| MARA (Our Framework)Model=GPT-4o, Input=top-32026.02 | 62.35 | |
| MARA (Our Framework)Model=GPT-4o, Input=top-12026.02 | 58.93 | |
| VisRAGModel=GPT-4o, Input=top-32026.02 | 58.27 | |
| MARAModel=MiniCPM-V 2.62026.02 | 58.07 | |
| VisRAGModel=GPT-4o, Input=top-12026.02 | 55.32 | |
| MARA (Our Framework)Model=MiniCPM-V 2.6, Input=top-32026.02 | 52.2 | |
| MARA (Our Framework)Model=MiniCPM-V 2.6, Input=top-12026.02 | 51.86 | |
| VisRAGModel=MiniCPM-V 2.6, Input=top-32026.02 | 50.94 | |
| VisRAGModel=MiniCPM-V 2.6, Input=top-12026.02 | 49.91 | |
| VisRAGModel=MiniCPM-V 2.6, Input=top-102026.02 | 43.2 | |
| MARA (Our Framework)Model=MiniCPM-V 2.6, Input=top-102026.02 | 37.73 |