Stitch and Tell: A Structured Multimodal Data Augmentation Method for Spatial Understanding

TL;DR

This paper introduces SiTe, a simple data augmentation method that improves spatial understanding in vision-language models by stitching images and generating spatially-aware captions without extra annotations.

Contribution

It proposes a novel, annotation-free data augmentation technique called Stitch and Tell (SiTe) that enhances spatial reasoning in vision-language models by injecting structured spatial supervision.

Findings

SiTe improves spatial understanding tasks by over 4% in benchmark scores.

It maintains or enhances performance on general vision-language benchmarks.

The method is simple, plug-and-play, and does not require costly annotations or advanced models.

Abstract

Existing vision-language models often suffer from spatial hallucinations, i.e., generating incorrect descriptions about the relative positions of objects in an image. We argue that this problem mainly stems from the asymmetric properties between images and text. To enrich the spatial understanding ability of vision-language models, we propose a simple, annotation-free, plug-and-play method named $\text{Stitch and Tell}$ (abbreviated as SiTe), which injects structured spatial supervision into data. It constructs stitched image-text pairs by stitching images along a spatial axis and generating spatially-aware captions or question answer pairs based on the layout of stitched image, without relying on costly advanced models or human involvement. We evaluate SiTe across three architectures including LLaVA-v1.5-7B, LLaVA-Qwen2-1.5B and HALVA-7B, two training datasets, and eight benchmarks. Experiments show that SiTe improves spatial understanding tasks such as $\text{MME}_{\text{Position}}$ (+5.50%) and Spatial-MM (+4.19%), while maintaining or improving performance on general vision-language benchmarks including COCO-QA (+1.02%) and MMBench (+4.76%). Our findings suggest that explicitly injecting spatially-aware structure into training data offers an effective way to mitigate spatial hallucinations and improve spatial understanding, while preserving general vision-language capabilities.