Grounded Image Editing Request (GIER) dataset
收藏资源简介:
Grounded Image Editing Request (GIER)数据集是一个大规模的语言驱动图像编辑数据集,由罗切斯特大学和Adobe研究院共同创建。该数据集包含6179对独特的图像,每对图像都标注了五种语言请求、所有可能的编辑操作及其掩码。数据集中的图像和请求均来自真实的用户编辑需求,涵盖了多样化的编辑操作。创建过程中,研究人员从图像编辑请求网站爬取数据,并通过专家和众包平台进行数据增强和标注质量控制。GIER数据集旨在推动语言驱动图像编辑技术的发展,解决图像编辑中的自动化和可解释性问题,使图像编辑更加易于普通用户使用。
Grounded Image Editing Request (GIER) dataset is a large-scale language-driven image editing dataset co-created by the University of Rochester and Adobe Research. This dataset contains 6,179 unique image pairs, each annotated with five language requests, all possible editing operations and their corresponding masks. Both the images and requests in the dataset originate from real user editing needs, covering a diverse range of editing operations. During the dataset creation process, researchers crawled data from image editing request websites, and conducted data augmentation and annotation quality control via expert and crowdsourcing platforms. The GIER dataset aims to promote the development of language-driven image editing technologies, address the challenges of automation and interpretability in image editing, and make image editing more accessible to ordinary users.




