On the Interplay of Convolutional Padding and Adversarial Robustness

Paul Gavrikov, Janis Keuper
8/12/2023
cs.CVcs.AIcs.LG

No official code found. Best implementation: paulgavrikov/padding_robustness (2 stars, CC-BY-SA-4.0, last commit 2025-02-20).

Abstract

It is common practice to apply padding prior to convolution operations to preserve the resolution of feature-maps in Convolutional Neural Networks (CNN). While many alternatives exist, this is often achieved by adding a border of zeros around the inputs. In this work, we show that adversarial attacks often result in perturbation anomalies at the image boundaries, which are the areas where padding is used. Consequently, we aim to provide an analysis of the interplay between padding and adversarial attacks and seek an answer to the question of how different padding modes (or their absence) affect adversarial robustness in various scenarios.

View on arXivView PDF

Code Implementations(1)

paulgavrikov/padding_robustnessCommunity100%
CC-BY-SA-4.0

Official repository for the ICCVW2023 paper "On the Interplay of Convolutional Padding and Adversarial Robustness".

21PythonOct 5, 20231 years agoCC-BY-SA-4.0
adversarial-attackscomputer-visionconvolutional-neural-networkspaddingrobustness

Similar Papers

ExploreADV: Towards exploratory attack for Neural Networks

84% match2023

Investigating and unmasking feature-level vulnerabilities of CNNs to adversarial perturbations

84% match2024

Imperceptible Adversarial Attack on Deep Neural Networks from Image Boundary

84% match2023

Defending Adversarial Patches via Joint Region Localizing and Inpainting

84% match2023

PadChannel: Improving CNN Performance through Explicit Padding Encoding

83% match2023

Investigating Human-Identifiable Features Hidden in Adversarial Perturbations

83% match2023

Cite this paper

@article{gavrikov2023interplay,
  title  = {On the Interplay of Convolutional Padding and Adversarial Robustness},
  author = {Paul Gavrikov and Janis Keuper},
  year   = {2023},
  eprint = {2308.06612},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2308.06612v1}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2308.06612.svg)](https://codeofpaper.com/paper/2308.06612)

Discussion