MAGICBRUSH

MAGICBRUSH是由俄亥俄州立大学创建的大规模手动标注数据集，专注于真实图像的指令引导编辑。该数据集包含超过10,000个手动标注的三元组（源图像、指令、目标图像），支持单轮和多轮的指令引导编辑，适用于训练大规模文本引导图像编辑模型。数据集的创建过程严格，通过DALL-E 2平台进行图像合成，确保图像质量。MAGICBRUSH的应用领域广泛，旨在通过自然语言指令简化图像编辑过程，提高编辑效率和质量，适用于摄影、广告和社交媒体等多个领域。

MAGICBRUSH is a large-scale manually annotated dataset developed by The Ohio State University, focusing on instruction-guided editing of real-world images. This dataset contains over 10,000 manually annotated triplets (source image, instruction, target image), supports single-turn and multi-turn instruction-guided editing, and is suitable for training large-scale text-guided image editing models. The dataset was built through a rigorous process, with image synthesis conducted via the DALL-E 2 platform to ensure image quality. MAGICBRUSH has a wide range of application scenarios, aiming to simplify the image editing workflow via natural language instructions, improve editing efficiency and quality, and is applicable to multiple fields such as photography, advertising and social media.