e-ViL
收藏资源简介:
e-ViL是一个大型数据集,由牛津大学计算机科学系创建,旨在为视觉-语言任务提供自然语言解释的基准。该数据集包含超过430,000个实例,每个实例包括图像、自然语言假设、分类标签和解释。数据集的创建过程涉及多个步骤,包括手动重新标记和使用多种过滤方法来提高数据质量。e-ViL数据集主要用于评估和比较不同模型在视觉-语言任务中生成自然语言解释的能力,特别是在需要复杂推理和常识知识的场景中。
e-ViL is a large-scale dataset developed by the Department of Computer Science at the University of Oxford, serving as a benchmark for providing natural language explanations in vision-language tasks. This dataset includes over 430,000 instances, each composed of an image, a natural language hypothesis, a classification label, and an explanation. The dataset construction process involves multiple steps, including manual relabeling and the use of various filtering methods to improve data quality. The e-ViL dataset is mainly utilized to evaluate and compare the ability of different models to generate natural language explanations in vision-language tasks, especially in scenarios requiring complex reasoning and commonsense knowledge.




