This paper studies ensembling in the era of Large Vision-Language Models (LVLMs). Ensembling is a classical method to combine different models to get increased performance. In the recent work on Encyclopedic-VQA the authors examine a wide variety of models to solve their task: from vanilla LVLMs, to models including the caption as extra context, to models augmented with Lens-based retrieval of Wikipedia pages. Intuitively these models are highly complementary, which should make them ideal for ensembling. Indeed, an oracle experiment shows potential gains from 48.8% accuracy (the best single model) all the way up to 67% (best possible ensemble). So it is a trivial exercise to create an ensemble with substantial real gains. Or is it?

翻译：本文研究了大型视觉-语言模型（LVLMs）时代中的集成方法。集成是一种经典方法，通过组合不同模型来提升性能。在近期关于百科全书式视觉问答（Encyclopedic-VQA）的研究中，作者考察了多种模型以解决其任务：从原始LVLMs到将图像描述作为额外上下文的模型，再到基于Lens检索维基百科页面增强的模型。直觉上，这些模型高度互补，使其成为集成方法的理想候选。事实上，一项或然实验显示，潜在收益可从48.8%的准确率（最佳单模型）提升至67%（最佳可能集成）。因此，构建一个能带来实质性收益的集成似乎是轻而易举的。但事实果真如此吗？