InstructPix2Pix dataset
收藏资源简介:
InstructPix2Pix数据集是由加州大学圣塔巴巴拉分校的研究团队创建的,包含大量原始图像和对应的编辑图像对,以及编辑指令。该数据集通过结合预训练的文本到图像生成模型和大型语言模型GPT3生成编辑指令,但存在指令与图像编辑不匹配的问题。Instruct-CLIP方法对该数据集进行了优化,生成了超过12万条的精确且丰富的编辑指令,用于训练指导图像编辑模型,以提高其输出结果与指令的一致性。
The InstructPix2Pix dataset was created by a research team from the University of California, Santa Barbara. It contains a large corpus of paired original and edited images, along with corresponding editing instructions. Initially, the dataset generated editing instructions by combining a pre-trained text-to-image generation model and the large language model GPT-3, but it suffered from inconsistencies between the provided instructions and the actual performed image edits. The Instruct-CLIP method optimized this dataset, generating over 120,000 precise and rich editing instructions. These instructions are used to train guided image editing models to improve the alignment between the models' output results and the given instructions.




