wit543/sea-vqa

SEA-VQA是一个专门用于评估视觉问答（VQA）模型在处理东南亚文化特定内容时表现的数据集。该数据集包含来自八个东南亚国家的图像，这些图像来自联合国教科文组织文化遗产收藏，展示了该地区的各种文化文物、实践和遗产地。数据集旨在通过包含需要特定文化知识的问题，评估和提高VQA模型的文化意识和理解能力。

SEA-VQA is a dataset designed to evaluate the performance of Visual Question Answering (VQA) models on culturally specific content from Southeast Asia (SEA). This dataset includes images from eight Southeast Asian countries, curated from the UNESCO Cultural Heritage collection. The dataset aims to assess and improve the cultural awareness and understanding of VQA models by including questions that require specific cultural knowledge. The dataset contains 515 images and 1,999 questions, each paired with an image and requiring understanding of cultural specifics to answer correctly. The questions are in multiple-choice format with four possible answers each. The creation of the dataset was aimed at addressing the lack of cultural specificity in existing VQA datasets, particularly focusing on Southeast Asia. The images and questions were generated using GPT-4 and verified by graduate students from Southeast Asia with expertise in computer vision and natural language processing.