Journey Before Destination: On the importance of Visual Faithfulness in Slow Thinking

TL;DR

This paper emphasizes the importance of visual faithfulness in reasoning chains generated by vision-language models, proposing a new evaluation metric and self-reflection method to improve the reliability of multimodal reasoning.

Contribution

It introduces a novel metric for assessing visual faithfulness in reasoning chains and a self-reflection technique to detect and regenerate unfaithful perception steps without training.

Findings

Reduces unfaithful perception rate in reasoning chains

Maintains final-answer accuracy while improving faithfulness

Enhances reliability of multimodal reasoning models

Abstract

Reasoning-augmented vision language models (VLMs) generate explicit chains of thought that promise greater capability and transparency but also introduce new failure modes: models may reach correct answers via visually unfaithful intermediate steps, or reason faithfully yet fail on the final prediction. Standard evaluations that only measure final-answer accuracy cannot distinguish these behaviors. We introduce the visual faithfulness of reasoning chains as a distinct evaluation dimension, focusing on whether the perception steps of a reasoning chain are grounded in the image. We propose a training- and reference-free framework that decomposes chains into perception versus reasoning steps and uses off-the-shelf VLM judges for step-level faithfulness, additionally verifying this approach through a human meta-evaluation. Building on this metric, we present a lightweight self-reflection procedure that detects and locally regenerates unfaithful perception steps without any training. Across multiple reasoning-trained VLMs and perception-heavy benchmarks, our method reduces Unfaithful Perception Rate while preserving final-answer accuracy, improving the reliability of multimodal reasoning.