Does It Fail to See or Fail to Know? Attributing Errors in Vision-Language Models
2026-07-06 • Computer Vision and Pattern Recognition
Computer Vision and Pattern RecognitionComputation and Language
AI summaryⓘ
The authors studied how vision-language models (VLMs) make mistakes when answering questions about images, especially when the answer needs extra knowledge not visible in the image. They showed that errors come from different causes like problems recognizing objects or failing to retrieve knowledge after recognition. Their main discovery is that it’s possible to predict these types of errors before the model gives an answer by examining internal signals during processing. This early warning helps know why the model might be uncertain, which could guide fixes like improving image clarity or adding external information.
Vision-Language ModelsVisual Question AnsweringUncertainty QuantificationEntity RecognitionKnowledge RetrievalPre-generation SignalsModel Failure ModesHidden StatesVisual TokensImage Repair
Authors
Khang Nhat Hoang Vo, Artem Vazhentsev, Artem Shelmanov, Timothy Baldwin, Yova Kementchedjhieva
Abstract
Vision-language models (VLMs) perform well on visual question answering with high-quality images but struggle when questions require knowledge beyond what is clearly and directly visible. In such settings, uncertainty quantification should not only indicate whether the model is likely to fail but also diagnose why it is uncertain, across dimensions such as perception, entity recognition, and knowledge retrieval. While prior work has focused on individual failure modes in isolation or treated incorrect answers as monolithic failures, we propose a unified framework for disentangling these failure modes and investigate whether pre-generation signals can predict these failure sources. Across a range of datasets and model families, we find a consistent pattern in VLM errors: some failures arise from visual or recognition bottlenecks, while others persist after the relevant entity is identified. Our main finding is that these failure sources can be predicted before decoding: recognition-related failures are best captured by visual-token representations, while failures that remain after recognition are better captured by prompt-conditioned hidden states. This pre-generation signal enables efficient failure-source prediction before the model produces an answer, allowing uncertain cases to be routed to targeted interventions such as image repair, entity recognition support, or external retrieval.