COSOCO

该数据集名为COSOCO，包含了3364张大型的RGB图像，这些图像展示了良性及被篡改的软件容器。该数据集旨在支持在容器化环境中，基于机器学习技术进行恶意软件的检测。数据集按照恶意软件类别进行了分层抽样，以70:10:20的比例划分为训练集、验证集和测试集。此外，数据集中还包含了指示恶意软件在字节级别所做更改的遮罩。该数据集涵盖了来自1297个独特软件包和495个独特恶意软件样本的图像，这些样本跨越了10个不同的恶意软件类别，其任务是针对Docker容器中的恶意软件进行检测。This dataset, named COSOCO, contains 3,364 large RGB images depicting benign and tampered software containers. It is developed to support malware detection using machine learning techniques in containerized environments. The dataset undergoes stratified sampling based on malware categories, and is split into training, validation, and test sets with a 70:10:20 ratio. Additionally, the dataset also includes masks that indicate byte-level modifications made by malware. The images are derived from 1,297 unique software packages and 495 unique malware samples across 10 distinct malware categories, with the core objective of this dataset being malware detection within Docker containers.

搜集汇总

背景与挑战

背景概述

COSOCO是一个合成数据集，包含3364张图像，用于表示良性和恶意软件受损的软件容器，这些图像通过字节到像素工具从Docker容器转换而来。该数据集旨在支持图像分类任务以识别受损容器，以及图像分割任务以定位容器内被恶意软件修改的文件和区域，每个受损实例都附带一个掩码图像来标记受影响的像素。