Vision language models are unreliable at trivial spatial cognition
This highlights a critical reliability issue for VLMs in real-world applications, making it an incremental but important contribution to understanding model limitations.
The paper tackled the problem of vision language models (VLMs) being unreliable at trivial spatial cognition tasks, such as recognizing object positions, and found that performance degrades with minor prompt variations, revealing limitations in their reasoning abilities.
Vision language models (VLMs) are designed to extract relevant visuospatial information from images. Some research suggests that VLMs can exhibit humanlike scene understanding, while other investigations reveal difficulties in their ability to process relational information. To achieve widespread applicability, VLMs must perform reliably, yielding comparable competence across a wide variety of related tasks. We sought to test how reliable these architectures are at engaging in trivial spatial cognition, e.g., recognizing whether one object is left of another in an uncluttered scene. We developed a benchmark dataset -- TableTest -- whose images depict 3D scenes of objects arranged on a table, and used it to evaluate state-of-the-art VLMs. Results show that performance could be degraded by minor variations of prompts that use logically equivalent descriptions. These analyses suggest limitations in how VLMs may reason about spatial relations in real-world applications. They also reveal novel opportunities for bolstering image caption corpora for more efficient training and testing.