CultureVLM: Characterizing and Improving Cultural Understanding of Vision-Language Models for over 100 Countries
Jun 11, 2025·
,,,,,,,·
1 min read
Shudong Liu
Yiqiao Jin
Cheng Li
Derek F. Wong
Qingsong Wen
Lichao Sun
Haipeng Chen
Xing Xie
Jindong Wang
Abstract
Vision-language models (VLMs) are deployed globally but exhibit substantial cultural blind spots. We introduce CultureVLM, a framework that characterizes and improves cultural understanding of VLMs across more than 100 countries. CultureVLM provides culturally-grounded benchmarks and training procedures that meaningfully reduce cultural bias while preserving general capability.
Type
Publication
CVPR 2025 VLMs4All Workshop
Abstract
Vision-language models (VLMs) are deployed globally but exhibit substantial cultural blind spots. We introduce CultureVLM, a framework that characterizes and improves cultural understanding of VLMs across more than 100 countries.
