LiveVQA-new
收藏资源简介:
LIVEVQA是一个用于评估多模态大型语言模型在处理最新现实世界视觉信息的能力的基准数据集。它包含超过107,000个样本,每个样本都是由一张近期图片(来自新闻、YouTube或学术论文)和多级别、经过仔细验证的问题答案集组成。数据通过自动化的多阶段管道收集和筛选,确保所有问题都是相关且具有挑战性的。LIVEVQA支持对最先进的多模态大型语言模型进行视觉推理和知识更新的研究。
LIVEVQA is a benchmark dataset for evaluating the capabilities of multimodal large language models in processing cutting-edge real-world visual information. It contains over 107,000 samples, each consisting of a recent image sourced from news, YouTube, or academic papers, paired with a multi-level, carefully validated question-answer set. The dataset is collected and filtered via an automated multi-stage pipeline, ensuring that all questions are relevant and challenging. LIVEVQA supports research on visual reasoning and knowledge updating for state-of-the-art multimodal large language models.




