V-ReasonBench: Toward Unified Reasoning Benchmark Suite for Video Generation Models

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Luo, Yang, Zhao, Xuanlei, Lin, Baijiong, Zhu, Lingting, Tang, Liyao, Liu, Yuqi, Chen, Ying-Cong, Qian, Shengju, Wang, Xin, You, Yang
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866911277567180800
author Luo, Yang
Zhao, Xuanlei
Lin, Baijiong
Zhu, Lingting
Tang, Liyao
Liu, Yuqi
Chen, Ying-Cong
Qian, Shengju
Wang, Xin
You, Yang
author_facet Luo, Yang
Zhao, Xuanlei
Lin, Baijiong
Zhu, Lingting
Tang, Liyao
Liu, Yuqi
Chen, Ying-Cong
Qian, Shengju
Wang, Xin
You, Yang
contents Recent progress in generative video models, such as Veo-3, has shown surprising zero-shot reasoning abilities, creating a growing need for systematic and reliable evaluation. We introduce V-ReasonBench, a benchmark designed to assess video reasoning across four key dimensions: structured problem-solving, spatial cognition, pattern-based inference, and physical dynamics. The benchmark is built from both synthetic and real-world image sequences and provides a diverse set of answer-verifiable tasks that are reproducible, scalable, and unambiguous. Evaluations of six state-of-the-art video models reveal clear dimension-wise differences, with strong variation in structured, spatial, pattern-based, and physical reasoning. We further compare video models with strong image models, analyze common hallucination behaviors, and study how video duration affects Chain-of-Frames reasoning. Overall, V-ReasonBench offers a unified and reproducible framework for measuring video reasoning and aims to support the development of models with more reliable, human-aligned reasoning skills.
format Preprint
id arxiv_https___arxiv_org_abs_2511_16668
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle V-ReasonBench: Toward Unified Reasoning Benchmark Suite for Video Generation Models
Luo, Yang
Zhao, Xuanlei
Lin, Baijiong
Zhu, Lingting
Tang, Liyao
Liu, Yuqi
Chen, Ying-Cong
Qian, Shengju
Wang, Xin
You, Yang
Computer Vision and Pattern Recognition
Recent progress in generative video models, such as Veo-3, has shown surprising zero-shot reasoning abilities, creating a growing need for systematic and reliable evaluation. We introduce V-ReasonBench, a benchmark designed to assess video reasoning across four key dimensions: structured problem-solving, spatial cognition, pattern-based inference, and physical dynamics. The benchmark is built from both synthetic and real-world image sequences and provides a diverse set of answer-verifiable tasks that are reproducible, scalable, and unambiguous. Evaluations of six state-of-the-art video models reveal clear dimension-wise differences, with strong variation in structured, spatial, pattern-based, and physical reasoning. We further compare video models with strong image models, analyze common hallucination behaviors, and study how video duration affects Chain-of-Frames reasoning. Overall, V-ReasonBench offers a unified and reproducible framework for measuring video reasoning and aims to support the development of models with more reliable, human-aligned reasoning skills.
title V-ReasonBench: Toward Unified Reasoning Benchmark Suite for Video Generation Models
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2511.16668