ReFocus: Visual Editing as a Chain of Thought for Structured Image Understanding

TL;DR

ReFocus introduces a framework that enables multimodal large language models to perform visual editing through code, improving structured image understanding tasks like interpreting tables and charts by enhancing visual reasoning capabilities.

Contribution

The paper presents ReFocus, a novel method allowing LLMs to generate visual edits via code, which improves reasoning over structured images without adding extra information.

Findings

ReFocus improves task performance by 11.0% on tables and 6.8% on charts.

Visual chain-of-thought supervision outperforms standard QA data.

ReFocus's visual editing enhances reasoning without extra information.

Abstract

Structured image understanding, such as interpreting tables and charts, requires strategically refocusing across various structures and texts within an image, forming a reasoning sequence to arrive at the final answer. However, current multimodal large language models (LLMs) lack this multihop selective attention capability. In this work, we introduce ReFocus, a simple yet effective framework that equips multimodal LLMs with the ability to generate "visual thoughts" by performing visual editing on the input image through code, shifting and refining their visual focuses. Specifically, ReFocus enables multimodal LLMs to generate Python codes to call tools and modify the input image, sequentially drawing boxes, highlighting sections, and masking out areas, thereby enhancing the visual reasoning process. We experiment upon a wide range of structured image understanding tasks involving tables and charts. ReFocus largely improves performance on all tasks over GPT-4o without visual editing, yielding an average gain of 11.0% on table tasks and 6.8% on chart tasks. We present an in-depth analysis of the effects of different visual edits, and reasons why ReFocus can improve the performance without introducing additional information. Further, we collect a 14k training set using ReFocus, and prove that such visual chain-of-thought with intermediate information offers a better supervision than standard VQA data, reaching a 8.0% average gain over the same model trained with QA pairs and 2.6% over CoT.