Are We Using the Right Benchmark: An Evaluation Framework for Visual Token Compression Methods

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Liao, Chenfei, Wang, Wensong, Wen, Zichen, Zheng, Xu, Wang, Yiyu, He, Haocong, Lyu, Yuanhuiyi, Jiang, Lutao, Zou, Xin, Fu, Yuqian, Ren, Bin, Zhang, Linfeng, Hu, Xuming
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866915944013496320
author Liao, Chenfei
Wang, Wensong
Wen, Zichen
Zheng, Xu
Wang, Yiyu
He, Haocong
Lyu, Yuanhuiyi
Jiang, Lutao
Zou, Xin
Fu, Yuqian
Ren, Bin
Zhang, Linfeng
Hu, Xuming
author_facet Liao, Chenfei
Wang, Wensong
Wen, Zichen
Zheng, Xu
Wang, Yiyu
He, Haocong
Lyu, Yuanhuiyi
Jiang, Lutao
Zou, Xin
Fu, Yuqian
Ren, Bin
Zhang, Linfeng
Hu, Xuming
contents Recent efforts to accelerate inference in Multimodal Large Language Models (MLLMs) have largely focused on visual token compression. The effectiveness of these methods is commonly evaluated by measuring the accuracy drop on existing MLLM benchmarks before and after compression. However, these benchmarks are originally designed to assess general perception and reasoning abilities, rather than the specific challenges posed by visual token compression, leading to a fundamental task mismatch. In this work, we uncover a counterintuitive yet consistent phenomenon: simple image downsampling outperforms many advanced visual token compression methods across multiple widely used benchmarks. Through a comprehensive empirical study spanning eight popular benchmarks and multiple state-of-the-art compression techniques, we show that (i) current benchmarks contain substantial noise (task-irrelevant samples) for evaluating visual token compression, and (ii) downsampling can act as an effective data filter that distinguishes between simple and difficult samples with respect to compression sensitivity. Motivated by these findings, we propose VTC-Bench, an evaluation framework that explicitly leverages downsampling as a discriminator to denoise existing benchmarks, enabling a fairer and more meaningful additional assessment of visual token compression methods.
format Preprint
id arxiv_https___arxiv_org_abs_2510_07143
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Are We Using the Right Benchmark: An Evaluation Framework for Visual Token Compression Methods
Liao, Chenfei
Wang, Wensong
Wen, Zichen
Zheng, Xu
Wang, Yiyu
He, Haocong
Lyu, Yuanhuiyi
Jiang, Lutao
Zou, Xin
Fu, Yuqian
Ren, Bin
Zhang, Linfeng
Hu, Xuming
Computer Vision and Pattern Recognition
Recent efforts to accelerate inference in Multimodal Large Language Models (MLLMs) have largely focused on visual token compression. The effectiveness of these methods is commonly evaluated by measuring the accuracy drop on existing MLLM benchmarks before and after compression. However, these benchmarks are originally designed to assess general perception and reasoning abilities, rather than the specific challenges posed by visual token compression, leading to a fundamental task mismatch. In this work, we uncover a counterintuitive yet consistent phenomenon: simple image downsampling outperforms many advanced visual token compression methods across multiple widely used benchmarks. Through a comprehensive empirical study spanning eight popular benchmarks and multiple state-of-the-art compression techniques, we show that (i) current benchmarks contain substantial noise (task-irrelevant samples) for evaluating visual token compression, and (ii) downsampling can act as an effective data filter that distinguishes between simple and difficult samples with respect to compression sensitivity. Motivated by these findings, we propose VTC-Bench, an evaluation framework that explicitly leverages downsampling as a discriminator to denoise existing benchmarks, enabling a fairer and more meaningful additional assessment of visual token compression methods.
title Are We Using the Right Benchmark: An Evaluation Framework for Visual Token Compression Methods
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2510.07143