Joint vision-language models have shown great performance over a diverse set
of tasks. However, little is known about their limitations, as the high
dimensional space learned by these models makes it difficult to identify
semantic errors. Recent work has addressed this problem by designing highly
controlled probing task benchmarks. Our paper introduces a more scalable
solution that relies on already annotated benchmarks. Our method consists of
extracting a large set of diverse features from a vision-language benchmark and
measuring their correlation with the output of the target model. We confirm
previous findings that CLIP behaves like a bag of words model and performs
better with nouns and verbs; we also uncover novel insights such as CLIP
getting confused by concrete words. Our framework is available at
this https URL and can be used with other
multimodal models and benchmarks.

本文介绍了一种更具可伸缩性的方法，其根据从视觉 - 语言基准中提取的大量多样化特征，并测量它们与目标模型输出的相关性。通过该方法，作者确认了之前发现的 CLIP 表现类似于词袋模型，并且在名词和动词上表现更好；作者还发现 CLIP 在处理具体词语时会出现混淆。该框架可用于其他多模态模型和基准测试。