DIFFUSIONDB
收藏资源简介:
DIFFUSIONDB是由佐治亚理工学院创建的第一个大规模文本到图像提示数据集,总容量达6.5TB,包含1400万张由Stable Diffusion生成的图像和180万个独特提示,以及由真实用户指定的超参数。该数据集通过收集Stable Diffusion公共Discord服务器上的图像构建,旨在帮助研究人员理解提示与生成模型之间的交互,检测深度伪造,并设计人机交互工具以更轻松地使用这些模型。DIFFUSIONDB的应用领域包括提示工程、深度伪造检测和大型生成模型的理解。
DIFFUSIONDB is the first large-scale text-to-image prompt dataset developed by the Georgia Institute of Technology. With a total capacity of 6.5 TB, it contains 14 million images generated by Stable Diffusion, 1.8 million unique prompts, and hyperparameters specified by real users. This dataset is curated by collecting images from public Stable Diffusion Discord servers, and its core objectives are to help researchers comprehend the interaction between prompts and generative models, detect deepfakes, and develop human-computer interaction tools to facilitate easier utilization of these models. The application areas of DIFFUSIONDB cover prompt engineering, deepfake detection, and the understanding of large generative models.

- 1DiffusionDB: A Large-scale Prompt Gallery Dataset for Text-to-Image Generative Models佐治亚理工学院 · 2023年



