Towards Robust 2D Convolution for Reliable Visual Recognition

TL;DR

This paper introduces RConv-MK, a novel convolutional module designed to enhance robustness against image corruptions and adversarial attacks in CNNs by using learnable kernels and adaptive noise removal, improving reliable visual recognition.

Contribution

The paper proposes RConv-MK, a new convolutional building block that improves robustness of CNN features through multi-scale kernels and adaptive soft thresholding, addressing vulnerability to corruptions.

Findings

RConv-MK improves robustness on corrupted images.

It enhances resilience against adversarial samples.

Experimental results validate effectiveness across various scenarios.

Abstract

2D convolution (Conv2d), which is responsible for extracting features from the input image, is one of the key modules of a convolutional neural network (CNN). However, Conv2d is vulnerable to image corruptions and adversarial samples. It is an important yet rarely investigated problem that whether we can design a more robust alternative of Conv2d for more reliable feature extraction. In this paper, inspired by the recently developed learnable sparse transform that learns to convert the CNN features into a compact and sparse latent space, we design a novel building block, denoted by RConv-MK, to strengthen the robustness of extracted convolutional features. Our method leverages a set of learnable kernels of different sizes to extract features at different frequencies and employs a normalized soft thresholding operator to adaptively remove noises and trivial features at different corruption levels. Extensive experiments on clean images, corrupted images as well as adversarial samples validate the effectiveness of the proposed robust module for reliable visual recognition. The source codes are enclosed in the submission.