Mitigating Hallucinations in Multimodal Spatial Relations through Constraint-Aware Prompting

TL;DR

This paper introduces a constraint-aware prompting framework that reduces spatial relation hallucinations in large vision-language models by enforcing bidirectional and transitivity constraints, leading to more coherent spatial predictions.

Contribution

The paper proposes a novel constraint-aware prompting method with bidirectional and transitivity constraints to mitigate spatial hallucinations in LVLMs.

Findings

Improved spatial relation accuracy on three datasets

Enhanced consistency in object relation predictions

Systematic analysis of constraint effectiveness

Abstract

Spatial relation hallucinations pose a persistent challenge in large vision-language models (LVLMs), leading to generate incorrect predictions about object positions and spatial configurations within an image. To address this issue, we propose a constraint-aware prompting framework designed to reduce spatial relation hallucinations. Specifically, we introduce two types of constraints: (1) bidirectional constraint, which ensures consistency in pairwise object relations, and (2) transitivity constraint, which enforces relational dependence across multiple objects. By incorporating these constraints, LVLMs can produce more spatially coherent and consistent outputs. We evaluate our method on three widely-used spatial relation datasets, demonstrating performance improvements over existing approaches. Additionally, a systematic analysis of various bidirectional relation analysis choices and transitivity reference selections highlights greater possibilities of our methods in incorporating constraints to mitigate spatial relation hallucinations.