Thinking with Images via Self-Calling Agent

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Yang, Wenxi, Zhao, Yuzhong, Wan, Fang, Ye, Qixiang
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908705245626368
author Yang, Wenxi
Zhao, Yuzhong
Wan, Fang
Ye, Qixiang
author_facet Yang, Wenxi
Zhao, Yuzhong
Wan, Fang
Ye, Qixiang
contents Thinking-with-images paradigms have showcased remarkable visual reasoning capability by integrating visual information as dynamic elements into the Chain-of-Thought (CoT). However, optimizing interleaved multimodal CoT (iMCoT) through reinforcement learning remains challenging, as it relies on scarce high-quality reasoning data. In this study, we propose Self-Calling Chain-of-Thought (sCoT), a novel visual reasoning paradigm that reformulates iMCoT as a language-only CoT with self-calling. Specifically, a main agent decomposes the complex visual reasoning task to atomic subtasks and invokes its virtual replicas, i.e. parameter-sharing subagents, to solve them in isolated context. sCoT enjoys substantial training effectiveness and efficiency, as it requires no explicit interleaving between modalities. sCoT employs group-relative policy optimization to reinforce effective reasoning behavior to enhance optimization. Experiments on HR-Bench 4K show that sCoT improves the overall reasoning performance by up to $1.9\%$ with $\sim 75\%$ fewer GPU hours compared to strong baseline approaches. Code is available at https://github.com/YWenxi/think-with-images-through-self-calling.
format Preprint
id arxiv_https___arxiv_org_abs_2512_08511
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Thinking with Images via Self-Calling Agent
Yang, Wenxi
Zhao, Yuzhong
Wan, Fang
Ye, Qixiang
Computer Vision and Pattern Recognition
Thinking-with-images paradigms have showcased remarkable visual reasoning capability by integrating visual information as dynamic elements into the Chain-of-Thought (CoT). However, optimizing interleaved multimodal CoT (iMCoT) through reinforcement learning remains challenging, as it relies on scarce high-quality reasoning data. In this study, we propose Self-Calling Chain-of-Thought (sCoT), a novel visual reasoning paradigm that reformulates iMCoT as a language-only CoT with self-calling. Specifically, a main agent decomposes the complex visual reasoning task to atomic subtasks and invokes its virtual replicas, i.e. parameter-sharing subagents, to solve them in isolated context. sCoT enjoys substantial training effectiveness and efficiency, as it requires no explicit interleaving between modalities. sCoT employs group-relative policy optimization to reinforce effective reasoning behavior to enhance optimization. Experiments on HR-Bench 4K show that sCoT improves the overall reasoning performance by up to $1.9\%$ with $\sim 75\%$ fewer GPU hours compared to strong baseline approaches. Code is available at https://github.com/YWenxi/think-with-images-through-self-calling.
title Thinking with Images via Self-Calling Agent
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2512.08511