# Image-Based Malware Type Classification on MalNet-Image Tiny: Effects of Multi-Scale Fusion, Transfer Learning, Data Augmentation, and Schedule-Free Optimization

Source: [Image-Based Malware Type Classification on MalNet-Image Tiny: Effects of Multi-Scale Fusion, Transfer Learning, Data Augmentation, and Schedule-Free Optimization](<https://www.sota2.com/research/paper/69eaf8eb55a3c32af89f87ed>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-04-25T05:00:04.396Z

Authors: Ahmed A. Abouelkhaire, Waleed A. Yousef, Issa Traor

Published: 2026-04-22T23:45:44.000Z

## Abstract

This paper studies 43-class malware type classification on MalNet-Image Tiny, a public benchmark derived from Android APK files. The goal is to assess whether a compact image classifier benefits from four components evaluated in a controlled ablation: a feature pyramid network (FPN) for scale variation induced by resizing binaries of different lengths, ImageNet pretraining, lightweight augmentation through Mixup and TrivialAugment, and schedule-free AdamW optimization. All experiments use a ResNet18 backbone and the provided train/validation/test split. Reproducing the benchmark-style configuration yields macro-F1 (F1\_macro) of 0.6510, consistent with the reported baseline of approximately 0.65. Replacing the optimizer with schedule-free AdamW and using unweighted cross-entropy increases F1\_macro to 0.6535 in 10 epochs, compared with 96 epochs for the reproduced baseline. The best configuration combines pretraining, Mixup, TrivialAugment, and FPN, reaching F1\_macro=0.6927, P\_macro=0.7707, AUC\_macro=0.9556, and L\_test=0.8536. The ablation indicates that the largest gains in F1\_macro arise from pretraining and augmentation, whereas FPN mainly improves P\_macro, AUC\_macro, and L\_test in the strongest configuration.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2604.21153>)

[PDF](<https://arxiv.org/pdf/2604.21153.pdf>)
