Synbols
收藏arXiv2020-11-05 更新2024-06-21 收录
下载链接:
https://github.com/ElementAI/synbols
下载链接
链接失效反馈官方服务:
资源简介:
Synbols是由Element AI开发的一个用于快速生成具有丰富潜在特征的低分辨率图像数据集的工具。该数据集利用Unicode标准中的大量符号和开源社区提供的多种艺术字体,通过Pycairo这一2D矢量图形库进行渲染。Synbols的高级接口支持快速生成新的数据分布,包括各种纹理和遮挡类型。该数据集广泛应用于机器学习领域,特别是在监督学习、主动学习、分布外泛化、无监督表示学习和对象计数等设置中,以揭示标准学习算法的局限性和缺陷。
Synbols is a tool developed by Element AI for rapidly generating low-resolution image datasets with rich latent features. This dataset leverages a large number of symbols from the Unicode standard and various artistic fonts provided by the open-source community, and renders images via Pycairo, a 2D vector graphics library. The advanced interface of Synbols supports rapid generation of novel data distributions, including various texture and occlusion types. This dataset is widely applied in the field of machine learning, particularly in scenarios such as supervised learning, active learning, out-of-distribution generalization, unsupervised representation learning, and object counting, to reveal the limitations and flaws of standard learning algorithms.
提供机构:
Element AI
创建时间:
2020-09-14



