SmoGVLM: A Small, Graph-enhanced Vision-Language Model

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Mondal, Debjyoti, Singh, Rituraj, Panda, Subhadarshi
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866914485271265280
author Mondal, Debjyoti
Singh, Rituraj
Panda, Subhadarshi
author_facet Mondal, Debjyoti
Singh, Rituraj
Panda, Subhadarshi
contents Large vision-language models (VLMs) achieve strong performance on multimodal tasks but often suffer from hallucination and poor grounding in knowledge-intensive reasoning. We propose SmoGVLM, a small, graph-enhanced VLM that integrates structured knowledge with visual and textual modalities, using Graph Neural Networks. We investigate the effects of our method across a range of model sizes, from tiny (1.3B) to large (13B) models. Our results demonstrate that, when trained using our approach, a small model can achieve performance gains upto 16.24%, and surpass its larger counterparts, outperforming larger VLMs and strong fine-tuned baselines. These results highlight the potential of structured knowledge augmentation for efficient, smaller-scale multimodal reasoning systems.
format Preprint
id arxiv_https___arxiv_org_abs_2604_16517
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle SmoGVLM: A Small, Graph-enhanced Vision-Language Model
Mondal, Debjyoti
Singh, Rituraj
Panda, Subhadarshi
Computer Vision and Pattern Recognition
Computation and Language
Large vision-language models (VLMs) achieve strong performance on multimodal tasks but often suffer from hallucination and poor grounding in knowledge-intensive reasoning. We propose SmoGVLM, a small, graph-enhanced VLM that integrates structured knowledge with visual and textual modalities, using Graph Neural Networks. We investigate the effects of our method across a range of model sizes, from tiny (1.3B) to large (13B) models. Our results demonstrate that, when trained using our approach, a small model can achieve performance gains upto 16.24%, and surpass its larger counterparts, outperforming larger VLMs and strong fine-tuned baselines. These results highlight the potential of structured knowledge augmentation for efficient, smaller-scale multimodal reasoning systems.
title SmoGVLM: A Small, Graph-enhanced Vision-Language Model
topic Computer Vision and Pattern Recognition
Computation and Language
url https://arxiv.org/abs/2604.16517