Enhancing Spatial Reasoning in Vision-Language Models via Chain-of-Thought Prompting and Reinforcement Learning

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Ji, Binbin, Agrawal, Siddharth, Tang, Qiance, Wu, Yvonne
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911063114514432
author Ji, Binbin
Agrawal, Siddharth
Tang, Qiance
Wu, Yvonne
author_facet Ji, Binbin
Agrawal, Siddharth
Tang, Qiance
Wu, Yvonne
contents This study investigates the spatial reasoning capabilities of vision-language models (VLMs) through Chain-of-Thought (CoT) prompting and reinforcement learning. We begin by evaluating the impact of different prompting strategies and find that simple CoT formats, where the model generates a reasoning step before the answer, not only fail to help, but can even harm the model's original performance. In contrast, structured multi-stage prompting based on scene graphs (SceneGraph CoT) significantly improves spatial reasoning accuracy. Furthermore, to improve spatial reasoning ability, we fine-tune models using Group Relative Policy Optimization (GRPO) on the SAT dataset and evaluate their performance on CVBench. Compared to supervised fine-tuning (SFT), GRPO achieves higher accuracy on Pass@1 evaluations and demonstrates superior robustness under out-of-distribution (OOD) conditions. In particular, we find that SFT overfits to surface-level linguistic patterns and may degrade performance when test-time phrasing changes (e.g., from "closer to" to "farther from"). GRPO, on the other hand, generalizes more reliably and maintains stable performance under such shifts. Our findings provide insights into how reinforcement learning and structured prompting improve the spatial reasoning capabilities and generalization behavior of modern VLMs. All code is open source at: https://github.com/Yvonne511/spatial-vlm-investigator
format Preprint
id arxiv_https___arxiv_org_abs_2507_13362
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Enhancing Spatial Reasoning in Vision-Language Models via Chain-of-Thought Prompting and Reinforcement Learning
Ji, Binbin
Agrawal, Siddharth
Tang, Qiance
Wu, Yvonne
Computer Vision and Pattern Recognition
Artificial Intelligence
Computation and Language
I.2.10; I.4.8; I.2.6; I.2.7; I.5.4; I.5.1
This study investigates the spatial reasoning capabilities of vision-language models (VLMs) through Chain-of-Thought (CoT) prompting and reinforcement learning. We begin by evaluating the impact of different prompting strategies and find that simple CoT formats, where the model generates a reasoning step before the answer, not only fail to help, but can even harm the model's original performance. In contrast, structured multi-stage prompting based on scene graphs (SceneGraph CoT) significantly improves spatial reasoning accuracy. Furthermore, to improve spatial reasoning ability, we fine-tune models using Group Relative Policy Optimization (GRPO) on the SAT dataset and evaluate their performance on CVBench. Compared to supervised fine-tuning (SFT), GRPO achieves higher accuracy on Pass@1 evaluations and demonstrates superior robustness under out-of-distribution (OOD) conditions. In particular, we find that SFT overfits to surface-level linguistic patterns and may degrade performance when test-time phrasing changes (e.g., from "closer to" to "farther from"). GRPO, on the other hand, generalizes more reliably and maintains stable performance under such shifts. Our findings provide insights into how reinforcement learning and structured prompting improve the spatial reasoning capabilities and generalization behavior of modern VLMs. All code is open source at: https://github.com/Yvonne511/spatial-vlm-investigator
title Enhancing Spatial Reasoning in Vision-Language Models via Chain-of-Thought Prompting and Reinforcement Learning
topic Computer Vision and Pattern Recognition
Artificial Intelligence
Computation and Language
I.2.10; I.4.8; I.2.6; I.2.7; I.5.4; I.5.1
url https://arxiv.org/abs/2507.13362