Hidden in plain sight: VLMs overlook their visual representations
This work diagnoses failure modes in open-source VLMs, which is important for researchers developing multimodal AI systems, though it appears incremental as it focuses on evaluation rather than proposing new solutions.
The paper found that vision-language models (VLMs) perform substantially worse than their visual encoders on vision-centric tasks like depth estimation and correspondence, dropping to near-chance performance, due to their failure to effectively use accessible visual information and reliance on language priors.
Language provides a natural interface to specify and evaluate performance on visual tasks. To realize this possibility, vision language models (VLMs) must successfully integrate visual and linguistic information. Our work compares VLMs to a direct readout of their visual encoders to understand their ability to integrate across these modalities. Across a series of vision-centric benchmarks (e.g., depth estimation, correspondence), we find that VLMs perform substantially worse than their visual encoders, dropping to near-chance performance. We investigate these results through a series of analyses across the entire VLM: namely 1) the degradation of vision representations, 2) brittleness to task prompt, and 3) the language model's role in solving the task. We find that the bottleneck in performing these vision-centric tasks lies in this third category; VLMs are not effectively using visual information easily accessible throughout the entire model, and they inherit the language priors present in the LLM. Our work helps diagnose the failure modes of open-source VLMs, and presents a series of evaluations useful for future investigations into visual understanding within VLMs.