K-Viscuit
收藏资源简介:
K-Viscuit是由韩国科学技术院人工智能研究所创建的针对韩国文化的视觉与文化解释测试数据集。该数据集包含657条记录,涵盖了韩国日常生活中的十大概念,如食物、游戏、宗教等。创建过程中,采用了人机协作的半自动化管道,确保了数据的质量和文化的相关性。数据集主要用于评估和提升视觉语言模型在理解和响应韩国文化特定输入方面的能力,旨在解决现有模型在非西方文化环境中表现不佳的问题。
K-Viscuit is a visual and cultural interpretation test dataset targeting Korean culture, developed by the Artificial Intelligence Research Institute of the Korea Advanced Institute of Science and Technology (KAIST). This dataset contains 657 records covering ten major concepts in daily Korean life, such as food, games, religion and other related domains. During its development, a semi-automated human-machine collaborative pipeline was adopted to ensure data quality and cultural relevance. The dataset is primarily used to evaluate and enhance the capabilities of vision-language models (VLMs) in understanding and responding to culturally specific inputs related to South Korea, aiming to address the poor performance of existing models in non-Western cultural contexts.

- 1Evaluating Visual and Cultural Interpretation: The K-Viscuit Benchmark with Human-VLM Collaboration韩国科学技术院人工智能研究所 · 2024年



