关于 MMSearch 基准的实验设置

[wysqwysq]

您好，请问在 Table 4 MMSearch 基准上测试时，DeepEyes V2 的文本搜索引擎是通过 SerpAPI 调用的 Google Search 吗？
如果不是，考虑到不同搜索引擎返回的信息质量可能存在差异，请问是如何确保与 MMSearch-R1 进行公平对比的呢？

[JaaackHongggg]

MMSearchR1 adopts a complex pipeline consisting of Serper, Jina Reader and a summarization module for its search process. In contrast, our approach only leverages Serper, making it simpler than that of MMSearchR1.
Furthermore, MMSearchR1 does not open-source the complete code and prompts for its search pipeline, which prevents us from conducting an alignment experiment.