PARC: A Quantitative Framework Uncovering the Symmetries within Vision Language Models
This work addresses the instability of VLMs to prompt variations, which is crucial for their reliable application in real-world problems, though it is incremental as it builds on existing analysis methods for LLMs.
The paper tackles the problem of prompt sensitivity in vision language models (VLMs) by introducing PARC, a framework for analyzing which prompt variations VLMs are most sensitive to and which models are most robust, finding that VLMs inherit language prompt sensitivity in the vision domain and identifying InternVL2 models as the most robust among 22 evaluated.
Vision language models (VLMs) respond to user-crafted text prompts and visual inputs, and are applied to numerous real-world problems. VLMs integrate visual modalities with large language models (LLMs), which are well known to be prompt-sensitive. Hence, it is crucial to determine whether VLMs inherit this instability to varying prompts. We therefore investigate which prompt variations VLMs are most sensitive to and which VLMs are most agnostic to prompt variations. To this end, we introduce PARC (Prompt Analysis via Reliability and Calibration), a VLM prompt sensitivity analysis framework built on three pillars: (1) plausible prompt variations in both the language and vision domain, (2) a novel model reliability score with built-in guarantees, and (3) a calibration step that enables dataset- and prompt-spanning prompt variation analysis. Regarding prompt variations, PARC's evaluation shows that VLMs mirror LLM language prompt sensitivity in the vision domain, and most destructive variations change the expected answer. Regarding models, outstandingly robust VLMs among 22 evaluated models come from the InternVL2 family. We further find indications that prompt sensitivity is linked to training data. The code will be at https://github.com/NVlabs/PARC.