RecipeQA
收藏资源简介:
RecipeQA是一个针对烹饪食谱多模态理解的数据集,由哈塞特佩大学计算机视觉实验室创建。该数据集包含约20,000个食谱,每个食谱配有标题、描述和一组对齐的图像,总计超过36,000个自动生成的问题-答案对。数据集通过多种理解任务,如文本填空、视觉填空、视觉连贯性和视觉排序,评估机器对图像和文本的联合理解能力,以及对事件时序流程和程序知识的理解。RecipeQA旨在为机器理解系统提供一个挑战性的测试平台,并作为评估基准。
RecipeQA is a multimodal understanding dataset focused on cooking recipes, created by the Computer Vision Laboratory of Hacettepe University. This dataset contains approximately 20,000 recipes, each equipped with a title, description, and a set of aligned images, totaling more than 36,000 automatically generated question-answer pairs. It evaluates machines' joint comprehension of images and text, as well as their understanding of event temporal sequences and procedural knowledge, via a range of understanding tasks including text cloze, visual cloze, visual coherence, and visual ordering. RecipeQA aims to serve as a challenging testbed and evaluation benchmark for machine comprehension systems.




