Prompted Textures Dataset (PTD)
收藏资源简介:
Prompted Textures Dataset (PTD) 是由威斯康星大学麦迪逊分校的研究团队创建的一个大规模、高质量的合成纹理图像数据集。该数据集包含362,880张图像,涵盖56种不同的纹理类别。数据集的创建过程包括使用文本描述生成提示,通过Stable Diffusion模型生成和过滤图像,并进一步使用CLIP评分进行质量筛选。PTD旨在支持广泛的基于纹理的任务,如纹理学习、解释性和鲁棒性研究,解决现有纹理数据集规模小、多样性不足的问题。
Prompted Textures Dataset (PTD) is a large-scale, high-quality synthetic texture image dataset developed by a research team from the University of Wisconsin-Madison. This dataset contains 362,880 images spanning 56 distinct texture categories. The dataset construction workflow includes generating prompts from textual descriptions, generating and filtering images using the Stable Diffusion model, and conducting further quality screening via CLIP scoring. PTD is designed to support a wide range of texture-based tasks such as texture learning, interpretability and robustness research, addressing the shortcomings of small scale and insufficient diversity in existing texture datasets.

- 1On Synthetic Texture Datasets: Challenges, Creation, and Curation威斯康星大学麦迪逊分校 · 2024年



