CompareBench: A Benchmark for Visual Comparison Reasoning in Vision-Language Models

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Cai, Jie, Yang, Kangning, Fu, Lan, Ding, Jiaming, Li, Jinlong, Sun, Huiming, Xing, Daitao, Shen, Jinglin, Meng, Zibo
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866914206151868416
author Cai, Jie
Yang, Kangning
Fu, Lan
Ding, Jiaming
Li, Jinlong
Sun, Huiming
Xing, Daitao
Shen, Jinglin
Meng, Zibo
author_facet Cai, Jie
Yang, Kangning
Fu, Lan
Ding, Jiaming
Li, Jinlong
Sun, Huiming
Xing, Daitao
Shen, Jinglin
Meng, Zibo
contents We introduce CompareBench, a benchmark for evaluating visual comparison reasoning in vision-language models (VLMs), a fundamental yet understudied skill. CompareBench consists of 1000 QA pairs across four tasks: quantity (600), temporal (100), geometric (200), and spatial (100). It is derived from two auxiliary datasets that we constructed: TallyBench (2000 counting images with QA) and HistCaps (515 historical images with bilingual captions). We evaluate both closed-source APIs (OpenAI, Gemini, Claude) and open-source models (Qwen2.5-VL and Qwen3-VL series). Results show clear scaling trends but also reveal critical limitations: even the strongest models consistently fail at temporal ordering and spatial relations, and they often make mistakes in basic counting and geometric comparisons that are trivial for humans. These findings demonstrate that visual comparison remains a systematic blind spot for current VLMs. By providing controlled, diverse, and diagnostic evaluation, CompareBench establishes a foundation for advancing more reliable multimodal reasoning.
format Preprint
id arxiv_https___arxiv_org_abs_2509_22737
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle CompareBench: A Benchmark for Visual Comparison Reasoning in Vision-Language Models
Cai, Jie
Yang, Kangning
Fu, Lan
Ding, Jiaming
Li, Jinlong
Sun, Huiming
Xing, Daitao
Shen, Jinglin
Meng, Zibo
Computer Vision and Pattern Recognition
Artificial Intelligence
We introduce CompareBench, a benchmark for evaluating visual comparison reasoning in vision-language models (VLMs), a fundamental yet understudied skill. CompareBench consists of 1000 QA pairs across four tasks: quantity (600), temporal (100), geometric (200), and spatial (100). It is derived from two auxiliary datasets that we constructed: TallyBench (2000 counting images with QA) and HistCaps (515 historical images with bilingual captions). We evaluate both closed-source APIs (OpenAI, Gemini, Claude) and open-source models (Qwen2.5-VL and Qwen3-VL series). Results show clear scaling trends but also reveal critical limitations: even the strongest models consistently fail at temporal ordering and spatial relations, and they often make mistakes in basic counting and geometric comparisons that are trivial for humans. These findings demonstrate that visual comparison remains a systematic blind spot for current VLMs. By providing controlled, diverse, and diagnostic evaluation, CompareBench establishes a foundation for advancing more reliable multimodal reasoning.
title CompareBench: A Benchmark for Visual Comparison Reasoning in Vision-Language Models
topic Computer Vision and Pattern Recognition
Artificial Intelligence
url https://arxiv.org/abs/2509.22737