Figure9
Figure 9. Qualitative results on our proposed VQA benchmark. Our method enhances the accuracy of answers from three MLLMs across nearly all question categories. Human-annotated answers are provided for each test sample. Incorrect and correct answers are highlighted in red green, respectively. VQA: Visual Question Answering.
