DiffuGen

DiffuGen是由德雷塞尔大学计算机科学系开发的创新数据集，旨在利用稳定扩散模型高效生成高质量的标记图像数据集。该数据集通过结合无监督和监督两种标记技术，利用提示模板化和文本反转增强扩散模型的能力，实现图像的多样化生成和精确标记。数据集创建过程中，采用了预训练的稳定扩散模型，并通过文本到图像、图像到图像和修复等任务扩展图像多样性。DiffuGen的应用领域广泛，特别适用于需要高度真实感和多样性的机器学习和计算机视觉研究，如车辆事故场景的模拟。

DiffuGen is an innovative dataset developed by the Department of Computer Science, Drexel University. It is designed to efficiently generate high-quality labeled image datasets using Stable Diffusion models. This dataset integrates both unsupervised and supervised labeling techniques, leveraging prompt templating and textual inversion to enhance the capabilities of diffusion models, thereby enabling diverse image generation and accurate labeling. During the dataset construction process, pre-trained Stable Diffusion models are adopted, and image diversity is expanded through tasks such as text-to-image, image-to-image, and inpainting. DiffuGen has a wide range of application scenarios, and is particularly suitable for machine learning and computer vision research requiring high realism and diversity, such as the simulation of vehicle accident scenes.