CVQA: Culturally-Diverse Multilingual VQA

CVQA is a large-scale multimodal evaluation benchmark created to evaluate the cultural diversity and multilingual capabilities of Vision-Language Models (VLMs).

• Publication: NeurIPS 2024 (“CVQA: Culturally-diverse Multilingual Visual Question Answering Benchmark”).
• Coverage: Over 30 languages spanning multiple continents, featuring imagery and question-answer pairs curated by local native speakers and researchers worldwide.
• Key Finding: Highlights significant performance degradation in state-of-the-art multimodal models when queried on non-Western cultural concepts.