VIVA
收藏资源简介:
VIVA数据集由香港理工大学和凯斯西储大学联合创建,专注于基于视觉的情境决策,特别强调人类价值观的融入。该数据集包含1062张描绘多样化现实情境的图像,每张图像都附有手动标注的决策及其背后的人类价值观和理由。数据集的创建过程涉及图像收集、多步骤标注及质量验证。VIVA数据集主要应用于评估和提升视觉语言模型在复杂社会情境中的决策能力,特别是在考虑人类价值观的前提下,旨在推动人工智能向更具社会责任感和人类中心的方向发展。
The VIVA dataset was co-created by The Hong Kong Polytechnic University and Case Western Reserve University, focusing on vision-based contextual decision-making with a particular emphasis on the integration of human values. This dataset contains 1062 images depicting diverse real-world scenarios, each paired with manually annotated decisions, along with the underlying human values and justifications for those decisions. The construction of the VIVA dataset involves image collection, multi-step annotation, and quality validation. Primarily, the VIVA dataset is utilized to evaluate and enhance the decision-making capabilities of vision-language models in complex social contexts, especially under the premise of considering human values, aiming to promote the development of artificial intelligence towards a more socially responsible and human-centric direction.

- 1VIVA: A Benchmark for Vision-Grounded Decision-Making with Human Values香港理工大学 2凯斯西储大学 · 2024年



