MoodArchive
收藏资源简介:
MoodArchive数据集由宾夕法尼亚州立大学创建,是一个包含超过8百万张图片的庞大数据库,每张图片都配有由LLaVA生成的详细的层次情绪注释,并经过人类评估者的部分验证。这些图片涵盖了27种情绪和4种不同的情境,包括面部表情、自然风光、城市风光和物体类别。数据集的创建过程包括使用ChatGPT将抽象情绪分解为具体的描述符,然后从多个来源检索图像,并使用LLaVA-NeXT生成结构化注释。该数据集旨在解决现有图像编辑工具在情感驱动控制方面的不足,并为动画艺术家、电子商务平台和营销团队等领域的情感内容创作提供新的可能性。
The MoodArchive dataset, developed by Pennsylvania State University, is a large-scale database containing over 8 million images. Each image is paired with detailed hierarchical emotion annotations generated by LLaVA, with partial validation performed by human evaluators. The dataset covers 27 emotion categories and 4 distinct scenarios, including facial expressions, natural landscapes, urban scenery, and object categories. The dataset construction workflow involves using ChatGPT to decompose abstract emotions into concrete descriptive terms, retrieving images from multiple sources, and generating structured annotations via LLaVA-NeXT. This dataset is designed to address the limitations of existing image editing tools in emotion-driven control, and offers novel opportunities for emotional content creation across domains such as animation artists, e-commerce platforms, and marketing teams.



