TWO4TWO
收藏arXiv2021-05-07 更新2024-06-21 收录
下载链接:
https://github.com/mschuessler/two4two/
下载链接
链接失效反馈官方服务:
资源简介:
TWO4TWO数据集由魏泽堡研究院和柏林工业大学创建,专注于可解释机器学习的评估。该数据集包含两个3D抽象动物图像,通过简单参数控制生成,适用于算法和人类评估。数据集创建过程中,通过随机参数采样生成图像,支持定制化数据集的生成。该数据集主要用于解决机器学习模型解释性的问题,特别是在人类参与的评估实验中,以验证解释技术的有效性和模型的可理解性。
The TWO4TWO dataset was developed by the Weizmann Institute of Science and Technische Universität Berlin, with a core focus on the evaluation of explainable machine learning. This dataset contains two categories of 3D abstract animal images, which are generated using controlled simple parameters, and is applicable to both algorithmic and human evaluations. During the dataset construction process, images are generated via random parameter sampling, which enables the creation of customized datasets. This dataset is primarily designed to address the challenge of machine learning model interpretability, particularly in human-in-the-loop evaluation experiments, to validate the effectiveness of explanation techniques and the intelligibility of the models.
提供机构:
魏泽堡研究院,柏林工业大学
创建时间:
2021-05-07



