SceneAlign: Aligning Multimodal Reasoning to Scene Graphs in Complex Visual Scenes

TL;DR

SceneAlign enhances multimodal reasoning in complex visual scenes by aligning language models with scene graphs, reducing hallucinations and improving answer accuracy through structured visual grounding and contrastive training.

Contribution

It introduces a novel framework that uses scene graphs for controllable structural interventions and contrastive training to improve reasoning faithfulness in multimodal models.

Findings

Consistently improves answer accuracy across seven benchmarks.

Reduces hallucinated entities and mis-grounded relations.

Enhances reasoning faithfulness in complex visual scenes.

Abstract

Multimodal large language models often struggle with faithful reasoning in complex visual scenes, where intricate entities and relations require precise visual grounding at each step. This reasoning unfaithfulness frequently manifests as hallucinated entities, mis-grounded relations, skipped steps, and over-specified reasoning. Existing preference-based approaches, typically relying on textual perturbations or answer-conditioned rationales, fail to address this challenge as they allow models to exploit language priors to bypass visual grounding. To address this, we propose SceneAlign, a framework that leverages scene graphs as structured visual information to perform controllable structural interventions. By identifying reasoning-critical nodes and perturbing them through four targeted strategies that mimic typical grounding failures, SceneAlign constructs hard negative rationales that remain linguistically plausible but are grounded in inaccurate visual facts. These contrastive pairs are used in Direct Preference Optimization to steer models toward fine-grained, structure-faithful reasoning. Across seven visual reasoning benchmarks, SceneAlign consistently improves answer accuracy and reasoning faithfulness, highlighting the effectiveness of grounding-aware alignment for multimodal reasoning.