VAGUE
收藏资源简介:
VAGUE数据集由延世大学和UC Berkeley的研究团队创建,旨在评估多模态模型在处理含糊不清的人类表达时的能力。该数据集包含3900条间接的人类话语及其对应的场景图像,适用于需要复杂语言和视觉推理的任务。数据集的创建过程包括生成提示-解决方案对,并设计了多选题和自由回答格式来评估模型。VAGUE的应用领域主要集中在提升模型对间接交流的理解能力,以实现更自然和直观的交互。
The VAGUE dataset was developed by research teams from Yonsei University and the University of California, Berkeley, with the goal of evaluating the performance of multimodal models when handling ambiguous human expressions. This dataset contains 3,900 indirect human utterances and their corresponding scene images, which is suitable for tasks requiring complex linguistic and visual reasoning. The creation process of the VAGUE dataset includes generating prompt-solution pairs, and multiple-choice and free-response formats are designed to evaluate models. The application scenarios of VAGUE mainly focus on enhancing the model's ability to understand indirect communication, so as to achieve more natural and intuitive interactions.




