Enhancing Alignment for Unified Multimodal Models via Semantically-Grounded Supervision

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Kim, Jiyeong, So, Yerim, Choi, Hyesong, Hwang, Uiwon, Min, Dongbo
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866910061223215104
author Kim, Jiyeong
So, Yerim
Choi, Hyesong
Hwang, Uiwon
Min, Dongbo
author_facet Kim, Jiyeong
So, Yerim
Choi, Hyesong
Hwang, Uiwon
Min, Dongbo
contents Unified Multimodal Models (UMMs) have emerged as a promising paradigm that integrates multimodal understanding and generation within a unified modeling framework. However, current generative training paradigms suffer from inherent limitations. We present Semantically-Grounded Supervision (SeGroS), a fine-tuning framework designed to resolve the granularity mismatch and supervisory redundancy in UMMs. At its core, we propose a novel visual grounding map to construct two complementary supervision signals. First, we formulate semantic Visual Hints to compensate for the sparsity of text prompts. Second, we generate a semantically-grounded Corrupted Input to explicitly enhance the supervision of masking-based UMMs by restricting the reconstruction loss to core text-aligned regions. Extensive evaluations on GenEval, DPGBench, and CompBench demonstrate that SeGroS significantly improves generation fidelity and cross-modal alignment across various UMM architectures.
format Preprint
id arxiv_https___arxiv_org_abs_2603_19807
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Enhancing Alignment for Unified Multimodal Models via Semantically-Grounded Supervision
Kim, Jiyeong
So, Yerim
Choi, Hyesong
Hwang, Uiwon
Min, Dongbo
Computer Vision and Pattern Recognition
Artificial Intelligence
Unified Multimodal Models (UMMs) have emerged as a promising paradigm that integrates multimodal understanding and generation within a unified modeling framework. However, current generative training paradigms suffer from inherent limitations. We present Semantically-Grounded Supervision (SeGroS), a fine-tuning framework designed to resolve the granularity mismatch and supervisory redundancy in UMMs. At its core, we propose a novel visual grounding map to construct two complementary supervision signals. First, we formulate semantic Visual Hints to compensate for the sparsity of text prompts. Second, we generate a semantically-grounded Corrupted Input to explicitly enhance the supervision of masking-based UMMs by restricting the reconstruction loss to core text-aligned regions. Extensive evaluations on GenEval, DPGBench, and CompBench demonstrate that SeGroS significantly improves generation fidelity and cross-modal alignment across various UMM architectures.
title Enhancing Alignment for Unified Multimodal Models via Semantically-Grounded Supervision
topic Computer Vision and Pattern Recognition
Artificial Intelligence
url https://arxiv.org/abs/2603.19807