CNN Filter Database
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五十亿个来自数百个已训练卷积神经网络（CNNs）的滤波器，覆盖了广泛的数据集、架构和视觉任务。该数据集使得研究者能够基于不同的元参数，包括数据类型、任务、架构和层深度，来研究卷积滤波器中的分布偏移。这项研究旨在分析已训练卷积滤波器中的分布偏移情况。
This dataset contains over 5 billion filters extracted from hundreds of pre-trained Convolutional Neural Networks (CNNs), spanning a diverse array of datasets, network architectures, and visual tasks. This dataset enables researchers to investigate distribution shifts within convolutional filters based on diverse meta-parameters, including data type, task, architecture, and layer depth. This study aims to analyze distribution shifts in pre-trained convolutional filters.
提供机构：
Paul Gavrikov
