Saved in:
Bibliographic Details
Main Authors: Zhang, Jianshu, Yao, Dongyu, Pi, Renjie, Liang, Paul Pu, Fung, Yi R.
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2502.12084
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908429735428096
author Zhang, Jianshu
Yao, Dongyu
Pi, Renjie
Liang, Paul Pu
Fung, Yi R.
author_facet Zhang, Jianshu
Yao, Dongyu
Pi, Renjie
Liang, Paul Pu
Fung, Yi R.
contents Visually linking matching cues is a crucial ability in daily life, such as identifying the same person in multiple photos based on their cues, even without knowing who they are. Despite the extensive knowledge that vision-language models (VLMs) possess, it remains largely unexplored whether they are capable of performing this fundamental task. To address this, we introduce \textbf{VLM2-Bench}, a benchmark designed to assess whether VLMs can Visually Link Matching cues, with 9 subtasks and over 3,000 test cases. Comprehensive evaluation across twelve VLMs, along with further analysis of various language-side and vision-side prompting methods, leads to a total of eight key findings. We identify critical challenges in models' ability to link visual cues, highlighting a significant performance gap. Based on these insights, we advocate for (i) enhancing core visual capabilities to improve adaptability and reduce reliance on prior knowledge, (ii) establishing clearer principles for integrating language-based reasoning in vision-centric tasks to prevent unnecessary biases, and (iii) shifting vision-text training paradigms toward fostering models' ability to independently structure and infer relationships among visual cues.
format Preprint
id arxiv_https___arxiv_org_abs_2502_12084
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle VLM2-Bench: A Closer Look at How Well VLMs Implicitly Link Explicit Matching Visual Cues
Zhang, Jianshu
Yao, Dongyu
Pi, Renjie
Liang, Paul Pu
Fung, Yi R.
Computation and Language
Visually linking matching cues is a crucial ability in daily life, such as identifying the same person in multiple photos based on their cues, even without knowing who they are. Despite the extensive knowledge that vision-language models (VLMs) possess, it remains largely unexplored whether they are capable of performing this fundamental task. To address this, we introduce \textbf{VLM2-Bench}, a benchmark designed to assess whether VLMs can Visually Link Matching cues, with 9 subtasks and over 3,000 test cases. Comprehensive evaluation across twelve VLMs, along with further analysis of various language-side and vision-side prompting methods, leads to a total of eight key findings. We identify critical challenges in models' ability to link visual cues, highlighting a significant performance gap. Based on these insights, we advocate for (i) enhancing core visual capabilities to improve adaptability and reduce reliance on prior knowledge, (ii) establishing clearer principles for integrating language-based reasoning in vision-centric tasks to prevent unnecessary biases, and (iii) shifting vision-text training paradigms toward fostering models' ability to independently structure and infer relationships among visual cues.
title VLM2-Bench: A Closer Look at How Well VLMs Implicitly Link Explicit Matching Visual Cues
topic Computation and Language
url https://arxiv.org/abs/2502.12084