Hierarchical Pre-Training of Vision Encoders with Large Language Models

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Lee, Eugene, Chang, Ting-Yu, Tsai, Jui-Huang, Diao, Jiajie, Lee, Chen-Yi
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866910091031085056
author Lee, Eugene
Chang, Ting-Yu
Tsai, Jui-Huang
Diao, Jiajie
Lee, Chen-Yi
author_facet Lee, Eugene
Chang, Ting-Yu
Tsai, Jui-Huang
Diao, Jiajie
Lee, Chen-Yi
contents The field of computer vision has experienced significant advancements through scalable vision encoders and multimodal pre-training frameworks. However, existing approaches often treat vision encoders and large language models (LLMs) as independent modules, limiting the integration of hierarchical visual features. In this work, we propose HIVE (Hierarchical Pre-Training of Vision Encoders), a novel framework that enhances vision-language alignment by introducing hierarchical cross-attention between the vision encoder and LLM. Unlike conventional methods that flatten image embeddings, HIVE enables structured feature fusion across multiple layers, improving gradient flow and representation learning. To optimize this interaction, we introduce a three-stage training strategy that progressively aligns the vision encoder with the LLM, ensuring stable optimization and effective multimodal fusion. Empirical evaluations demonstrate that HIVE achieves superior performance not only in image classification but also on various vision-language tasks, outperforming self-attention-based methods in benchmarks such as MME, GQA, OK-VQA, and ScienceQA. Our results highlight the benefits of hierarchical feature integration, paving the way for more efficient and expressive vision-language models.
format Preprint
id arxiv_https___arxiv_org_abs_2604_00086
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Hierarchical Pre-Training of Vision Encoders with Large Language Models
Lee, Eugene
Chang, Ting-Yu
Tsai, Jui-Huang
Diao, Jiajie
Lee, Chen-Yi
Computer Vision and Pattern Recognition
Artificial Intelligence
Computation and Language
Machine Learning
I.2; I.4; H.3
The field of computer vision has experienced significant advancements through scalable vision encoders and multimodal pre-training frameworks. However, existing approaches often treat vision encoders and large language models (LLMs) as independent modules, limiting the integration of hierarchical visual features. In this work, we propose HIVE (Hierarchical Pre-Training of Vision Encoders), a novel framework that enhances vision-language alignment by introducing hierarchical cross-attention between the vision encoder and LLM. Unlike conventional methods that flatten image embeddings, HIVE enables structured feature fusion across multiple layers, improving gradient flow and representation learning. To optimize this interaction, we introduce a three-stage training strategy that progressively aligns the vision encoder with the LLM, ensuring stable optimization and effective multimodal fusion. Empirical evaluations demonstrate that HIVE achieves superior performance not only in image classification but also on various vision-language tasks, outperforming self-attention-based methods in benchmarks such as MME, GQA, OK-VQA, and ScienceQA. Our results highlight the benefits of hierarchical feature integration, paving the way for more efficient and expressive vision-language models.
title Hierarchical Pre-Training of Vision Encoders with Large Language Models
topic Computer Vision and Pattern Recognition
Artificial Intelligence
Computation and Language
Machine Learning
I.2; I.4; H.3
url https://arxiv.org/abs/2604.00086