Saved in:
Bibliographic Details
Main Authors: Hu, Lexiang, Xue, Youze, Li, Dian, Liu, Gang, Lin, Zhouchen
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2602.05729
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915778008186880
author Hu, Lexiang
Xue, Youze
Li, Dian
Liu, Gang
Lin, Zhouchen
author_facet Hu, Lexiang
Xue, Youze
Li, Dian
Liu, Gang
Lin, Zhouchen
contents Multimodal embeddings serve as a bridge for aligning vision and language, with the two primary implementations -- CLIP-based and MLLM-based embedding models -- both limited to capturing only global semantic information. Although numerous studies have focused on fine-grained understanding, we observe that complex scenarios currently targeted by MLLM embeddings often involve a hybrid perceptual pattern of both global and fine-grained elements, thus necessitating a compatible fusion mechanism. In this paper, we propose Adaptive Global and Fine-grained perceptual Fusion for MLLM Embeddings (AGFF-Embed), a method that prompts the MLLM to generate multiple embeddings focusing on different dimensions of semantic information, which are then adaptively and smoothly aggregated. Furthermore, we adapt AGFF-Embed with the Explicit Gradient Amplification (EGA) technique to achieve in-batch hard negatives enhancement without requiring fine-grained editing of the dataset. Evaluation on the MMEB and MMVP-VLM benchmarks shows that AGFF-Embed comprehensively achieves state-of-the-art performance in both general and fine-grained understanding compared to other multimodal embedding models.
format Preprint
id arxiv_https___arxiv_org_abs_2602_05729
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Adaptive Global and Fine-Grained Perceptual Fusion for MLLM Embeddings Compatible with Hard Negative Amplification
Hu, Lexiang
Xue, Youze
Li, Dian
Liu, Gang
Lin, Zhouchen
Computer Vision and Pattern Recognition
Machine Learning
Multimodal embeddings serve as a bridge for aligning vision and language, with the two primary implementations -- CLIP-based and MLLM-based embedding models -- both limited to capturing only global semantic information. Although numerous studies have focused on fine-grained understanding, we observe that complex scenarios currently targeted by MLLM embeddings often involve a hybrid perceptual pattern of both global and fine-grained elements, thus necessitating a compatible fusion mechanism. In this paper, we propose Adaptive Global and Fine-grained perceptual Fusion for MLLM Embeddings (AGFF-Embed), a method that prompts the MLLM to generate multiple embeddings focusing on different dimensions of semantic information, which are then adaptively and smoothly aggregated. Furthermore, we adapt AGFF-Embed with the Explicit Gradient Amplification (EGA) technique to achieve in-batch hard negatives enhancement without requiring fine-grained editing of the dataset. Evaluation on the MMEB and MMVP-VLM benchmarks shows that AGFF-Embed comprehensively achieves state-of-the-art performance in both general and fine-grained understanding compared to other multimodal embedding models.
title Adaptive Global and Fine-Grained Perceptual Fusion for MLLM Embeddings Compatible with Hard Negative Amplification
topic Computer Vision and Pattern Recognition
Machine Learning
url https://arxiv.org/abs/2602.05729