TY - JOUR
T1 - Towards Better Certified Segmentation via Diffusion Models
AU - Laousy, Othmane
AU - Araujo, Alexandre
AU - Chassagnon, Guillaume
AU - Revel, Marie Pierre
AU - Garg, Siddharth
AU - Khorrami, Farshad
AU - Vakalopoulou, Maria
N1 - Funding Information:
This work was granted access to the HPC resources of IDRIS under the allocation 2023-AD011013308R1 made by GENCI.
Publisher Copyright:
© UAI 2023. All rights reserved.
PY - 2023
Y1 - 2023
N2 - 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.
AB - 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.
UR - http://www.scopus.com/inward/record.url?scp=85170097330&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85170097330&partnerID=8YFLogxK
M3 - Conference article
AN - SCOPUS:85170097330
SN - 2640-3498
VL - 216
SP - 1185
EP - 1195
JO - Proceedings of Machine Learning Research
JF - Proceedings of Machine Learning Research
T2 - 39th Conference on Uncertainty in Artificial Intelligence, UAI 2023
Y2 - 31 July 2023 through 4 August 2023
ER -