LC-Mis Dataset

该数据集专为探究文本到图像生成模型中潜在的Concept Misalignment问题而设计，其中包含272个高质量的概念对。在这些概念对中，有173对专门针对潜在的Concept Misalignment。该数据集规模宏大，包含2,495个基于句子的文本提示，进而生成了9,980张图像。其任务旨在评估并减少文本到图像生成过程中的潜在概念不一致问题。

This dataset is specifically designed to investigate the latent Concept Misalignment issue in text-to-image generation models, and it contains 272 high-quality concept pairs. Out of these pairs, 173 are specifically targeted at latent Concept Misalignment. This large-scale dataset includes 2,495 sentence-based text prompts, which are used to generate 9,980 images. The core task of this dataset is to evaluate and mitigate latent Concept Misalignment issues during the text-to-image generation process.