Enhancing Alignment for Unified Multimodal Models via Semantically-Grounded Supervision

TL;DR

This paper introduces Semantically-Grounded Supervision (SeGroS), a novel fine-tuning framework that enhances the alignment and generation quality of Unified Multimodal Models by addressing supervision limitations through visual grounding maps.

Contribution

SeGroS provides a new supervision method using visual grounding maps to improve multimodal model training, addressing granularity mismatch and supervisory redundancy issues.

Findings

Significant improvement in generation fidelity.

Enhanced cross-modal alignment across architectures.

Effective supervision signals via visual grounding maps.

Abstract

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.