CNN Filter DB
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/paulgavrikov/cnn-filter-db
下载链接
链接失效反馈官方服务：
资源简介：
该数据集包含了超过14亿个来自数百个已训练卷积神经网络（CNNs）的滤波器，特别关注了3x3卷积滤波核的性质。此外，该数据集被用于分析不同架构和任务中CNN滤波器的分布变化及其特性。其规模之大，涵盖了来自各种模型的超过14亿个滤波器，旨在对已训练的卷积滤波器及其鲁棒性进行深入分析。
This dataset contains over 1.4 billion filters sourced from hundreds of pre-trained Convolutional Neural Networks (CNNs), with a particular focus on the properties of 3x3 convolutional kernels. Furthermore, it is utilized to analyze the distribution shifts and inherent characteristics of CNN filters across various architectures and tasks. Boasting such a large scale with more than 1.4 billion filters collected from diverse models, this dataset aims to enable in-depth analyses of pre-trained convolutional filters and their robustness.
提供机构：
Paul Gavrikov
