TCC-Bench

该数据集是一个双语视觉问答（VQA）基准，旨在评估多语言预训练模型（MLLMs）对传统文化的理解能力。它融合了文化丰富、视觉多样的数据，并采用半自动化的提问生成流程，重点在于减少语言偏见，并通过人工审核确保数据质量。该数据集的任务是视觉问答。This dataset is a bilingual visual question answering (VQA) benchmark designed to evaluate the performance of multilingual pre-trained language models (MLLMs) in comprehending traditional culture. It integrates culturally rich and visually diverse datasets, and adopts a semi-automated question generation workflow, with core objectives including mitigating linguistic bias and ensuring data quality via manual review. The primary task of this dataset is visual question answering.