Salvato in:
Dettagli Bibliografici
Autori principali: Li, Yifan, Bao, Wentao, Ye, Botao, Tan, Zhen, Chen, Tianlong, Liu, Huan, Kong, Yu
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:https://arxiv.org/abs/2504.04024
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866913778085396480
author Li, Yifan
Bao, Wentao
Ye, Botao
Tan, Zhen
Chen, Tianlong
Liu, Huan
Kong, Yu
author_facet Li, Yifan
Bao, Wentao
Ye, Botao
Tan, Zhen
Chen, Tianlong
Liu, Huan
Kong, Yu
contents To effectively reduce the visual tokens in Visual Large Language Models (VLLMs), we propose a novel approach called Window Token Concatenation (WiCo). Specifically, we employ a sliding window to concatenate spatially adjacent visual tokens. However, directly concatenating these tokens may group diverse tokens into one, and thus obscure some fine details. To address this challenge, we propose fine-tuning the last few layers of the vision encoder to adaptively adjust the visual tokens, encouraging that those within the same window exhibit similar features. To further enhance the performance on fine-grained visual understanding tasks, we introduce WiCo+, which decomposes the visual tokens in later layers of the LLM. Such a design enjoys the merits of the large perception field of the LLM for fine-grained visual understanding while keeping a small number of visual tokens for efficient inference. We perform extensive experiments on both coarse- and fine-grained visual understanding tasks based on LLaVA-1.5 and Shikra, showing better performance compared with existing token reduction projectors. The code is available: https://github.com/JackYFL/WiCo.
format Preprint
id arxiv_https___arxiv_org_abs_2504_04024
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Window Token Concatenation for Efficient Visual Large Language Models
Li, Yifan
Bao, Wentao
Ye, Botao
Tan, Zhen
Chen, Tianlong
Liu, Huan
Kong, Yu
Computer Vision and Pattern Recognition
To effectively reduce the visual tokens in Visual Large Language Models (VLLMs), we propose a novel approach called Window Token Concatenation (WiCo). Specifically, we employ a sliding window to concatenate spatially adjacent visual tokens. However, directly concatenating these tokens may group diverse tokens into one, and thus obscure some fine details. To address this challenge, we propose fine-tuning the last few layers of the vision encoder to adaptively adjust the visual tokens, encouraging that those within the same window exhibit similar features. To further enhance the performance on fine-grained visual understanding tasks, we introduce WiCo+, which decomposes the visual tokens in later layers of the LLM. Such a design enjoys the merits of the large perception field of the LLM for fine-grained visual understanding while keeping a small number of visual tokens for efficient inference. We perform extensive experiments on both coarse- and fine-grained visual understanding tasks based on LLaVA-1.5 and Shikra, showing better performance compared with existing token reduction projectors. The code is available: https://github.com/JackYFL/WiCo.
title Window Token Concatenation for Efficient Visual Large Language Models
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2504.04024