Mal4Sight

Our Mal4Sight dataset contains a set of executable files converted into grayscale images for researching image-based malware detection. The dataset includes 16,019 samples, comprising 11,971 malware samples from 35 families and 4,048 benign samples. Each executable file is divided into distinct sections: .text, .data, other and header. Each section is turned into a grayscale image of 256 x 256 pixels with byte-level structural patterns. The dataset is organised into training, validation, and testing subsets with set splits for both malware and benign samples. Each family folder contains multiple sample folders, each folder representing a malware executable. Every sample folder contains four grayscale images. Benign files do not belong to any malware family. For this reason, the benign directory does not contain family-wise subfolders. Instead, it contains sample folders, each of which stores the four grayscale section images of a benign executable. The naming conventions for the section images for both malware and benign files in our dataset are: header.png, text.png, data.png, and .other.png. The train directory contains 8,291 malware samples and 2,833 benign samples. The val directory contains 1,833 malware samples and 607 benign samples. The test directory contains 1,847 malware samples and 608 benign samples. In total, the dataset contains 64,076 grayscale images generated from 16,019 samples, of which 11,971 are malware and 4,048 are benign files. All images are stored in PNG format to keep consistent quality. This dataset can be used to train and evaluate machine learning and deep learning models for malware detection. It also supports tasks such as feature extraction, pattern analysis, and comparisons of image-based methods across various malware families. A metadata.xlsx file is included with the dataset and contains detailed information about the images, including file_id, class, section, split, family, image_path, width, channels, dtype, preprocessing, and is_valid fields. The dataset also includes: A list of the names of all 35 malware families Raw numerical data for branch contribution analysis Source code for the Baseline CNN and ResNet training pipelines PE parsing code with a fallback mechanism included. For additional information, visit https://github.com/WAHIDUL-ISLAM/Malware_Detection_Using_Image_Processing