Decompose and Compare Consistency: Measuring VLMs' Answer Reliability via Task-Decomposition Consistency Comparison

TL;DR

This paper introduces DeCC, a novel method for assessing the reliability of VLM responses by comparing direct answers with decomposed sub-question answers, improving accuracy correlation across multiple tasks.

Contribution

DeCC is a new approach that measures VLM answer reliability through consistency comparison between direct and decomposed question reasoning.

Findings

DeCC outperforms existing methods in correlating with task accuracy.

DeCC is effective across six vision-language tasks and three VLMs.

The method reduces overconfidence and confirmation bias issues.

Abstract

Despite tremendous advancements, current state-of-the-art Vision-Language Models (VLMs) are still far from perfect. They tend to hallucinate and may generate biased responses. In such circumstances, having a way to assess the reliability of a given response generated by a VLM is quite useful. Existing methods, such as estimating uncertainty using answer likelihoods or prompt-based confidence generation, often suffer from overconfidence. Other methods use self-consistency comparison but are affected by confirmation biases. To alleviate these, we propose Decompose and Compare Consistency (DeCC) for reliability measurement. By comparing the consistency between the direct answer generated using the VLM's internal reasoning process, and the indirect answers obtained by decomposing the question into sub-questions and reasoning over the sub-answers produced by the VLM, DeCC measures the reliability of VLM's direct answer. Experiments across six vision-language tasks with three VLMs show DeCC's reliability estimation achieves better correlation with task accuracy compared to the existing methods.