Towards Better Certified Segmentation via Diffusion Models

TL;DR

This paper proposes a novel approach combining randomized smoothing and diffusion models to enhance the certified robustness of image segmentation, achieving significant accuracy improvements without extra training.

Contribution

It introduces a new method that improves segmentation robustness by integrating diffusion models with randomized smoothing, surpassing previous state-of-the-art results.

Findings

Achieved a 21-point mean accuracy improvement over prior methods.

Method is model-agnostic and requires no additional training.

Significantly enhances certified robustness on Pascal-Context and Cityscapes datasets.

Abstract

The robustness of image segmentation has been an important research topic in the past few years as segmentation models have reached production-level accuracy. However, like classification models, segmentation models can be vulnerable to adversarial perturbations, which hinders their use in critical-decision systems like healthcare or autonomous driving. Recently, randomized smoothing has been proposed to certify segmentation predictions by adding Gaussian noise to the input to obtain theoretical guarantees. However, this method exhibits a trade-off between the amount of added noise and the level of certification achieved. In this paper, we address the problem of certifying segmentation prediction using a combination of randomized smoothing and diffusion models. Our experiments show that combining randomized smoothing and diffusion models significantly improves certified robustness, with results indicating a mean improvement of 21 points in accuracy compared to previous state-of-the-art methods on Pascal-Context and Cityscapes public datasets. Our method is independent of the selected segmentation model and does not need any additional specialized training procedure.