CVQA: Culturally-diverseMultilingual VisualQuestionAnsweringBenchmark SupplementaryMaterial

For documenting CVQA, we use the datasheet for datasets introduced by Gebru et al.[1], which9 specify the motivation, composition, collection process, preprocessing, uses and distribution of a10 dataset. For what purpose was this dataset created? To overcome these issues, we introduce18 CVQA, a new benchmark designed to include culturally-driven images and questions from 2819 countries covering26languages and11scripts. Are there multiple types of instances (e.g., movies, users, and ratings; people and35 interactionsbetweenthem;nodesandedges)? Specifically,eachinstanceisa37 dictionary thatcontains: image, ID, Subset, Question, Translated Question, Options,38 Translated Options, Label, Category, Image Type, Image Source, License.