Torchattacks: A PyTorch Repository for Adversarial Attacks

TL;DR

Torchattacks consolidates a diverse set of adversarial attack implementations in PyTorch, providing practical tools for generating adversarial examples and assessing robustness. It standardizes attack interfaces via a common base class and supports composite attacks through MultiAttack, facilitating robust evaluation and adversarial training workflows. The collection covers FGSM, BIM, CW, PGD variants, EOT, TRADES, fast adversarial training variants, and momentum-based methods, with clear formulas and implementation details. This repository serves as a practical, extensible resource for researchers and engineers aiming to verify model robustness and experiment with adversarial training strategies in a PyTorch environment.

Abstract

Torchattacks is a PyTorch library that contains adversarial attacks to generate adversarial examples and to verify the robustness of deep learning models. The code can be found at https://github.com/Harry24k/adversarial-attacks-pytorch.