500 Images per Digit Dataset
收藏arXiv2025-09-30 收录
下载链接：
https://dagecc-challenge.github.io/icpr2024/
下载链接
链接失效反馈官方服务：
资源简介：
该数据集包含了每个数字500张由预训练的稳定扩散模型生成的综合图像，这些图像展示了数字在金属表面上凸起或雕刻且呈现出风化效果的样子。创建此数据集的目的是为了模拟风化数字图像，以提升模型训练的效果。规模上，每个数字均有500张图像，任务是对这些数字字符进行分类。
This dataset contains 500 synthetic images per digit, generated by a pretrained Stable Diffusion model. These images depict digits that are either raised or engraved on metallic surfaces, bearing a weathered appearance. The primary motivation for constructing this dataset is to simulate weathered digit images for enhancing model training performance. In terms of scale, each digit has 500 images, and the core classification task of this dataset is to categorize these digit characters.
提供机构：
Fraunhofer IIS DEAL
