Multi-Concept Personalization Dataset

该数据集是一组高质量的图像集合，这些图像来自多部电影，展示了众多角色。此外，数据集还包含了手动生成的多概念问答样本。该数据集分为训练集和测试集，其中训练集包含单一概念的图像，而测试集则包含了单一和多重概念的图像。在构建此数据集时，使用了GPT-4o生成问答对，并强调了对具有多重概念实例的关注，以增强其在现实世界的适用性。该数据集大约包含了1.6千张图像，其任务涵盖了视觉问答（VQA）以及多概念个性化。

This dataset is a high-quality image collection sourced from various films, showcasing numerous characters. Additionally, it includes manually generated multi-concept question-answering samples. The dataset is split into training and test subsets: the training set contains images associated with single concepts, while the test set covers images with both single and multiple concepts. During the dataset construction process, GPT-4o was utilized to generate question-answering pairs, with special emphasis on instances involving multiple concepts to enhance its real-world applicability. The dataset contains approximately 1.6 thousand images, and its supported tasks include visual question answering (VQA) and multi-concept personalization.