Mal4Sight: A Structured Four-Part Image Dataset for Malware and Benign Executables

The Mal4Sight dataset is a four-section image dataset constructed from both malware and benign Portable Executable (PE) files. Each executable is decomposed into four distinct sections: header, .text, .data and .rsrc, and each section is converted into a corresponding grayscale image. The dataset consists of 12,283 malware executables from 35 families and 5,974 benign executables, resulting in 49,132 malware images and 23,896 benign images, for a total of 73,028 grayscale images. Each executable is represented as a folder containing four section-wise images generated from the same file. The dataset is organised into training, validation, and testing splits, with 8,580 malware and 4,181 benign samples in training, 1,827 malware and 896 benign samples in validation, and 1,876 malware and 897 benign samples in testing. For implementation details and source code, visit https://github.com/WAHIDUL-ISLAM/Malware_Detection_Using_Deep_Learning