# RePAIR: Inverting Backdoor Attacks in Semantic Segmentation

DevFeed: [RePAIR: Inverting Backdoor Attacks in Semantic Segmentation](<https://devfeed.tech/articles/repair-inverting-backdoor-attacks-in-semantic-segmentation-103372.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.37015>)

Author: Arnold Brosch, Kira Maag

Published: 2026-10-28T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [Adversarial attacks](<https://devfeed.tech/topics/adversarial-attacks.md>), [backdoor](<https://devfeed.tech/topics/backdoor.md>), [Cybersecurity](<https://devfeed.tech/topics/cybersecurity.md>)

Tags: [adversarial](<https://devfeed.tech/tags/adversarial.md>), [architectures](<https://devfeed.tech/tags/architectures.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [attack](<https://devfeed.tech/tags/attack.md>), [attacks](<https://devfeed.tech/tags/attacks.md>), [backdoor](<https://devfeed.tech/tags/backdoor.md>), [classification](<https://devfeed.tech/tags/classification.md>), [computer-vision](<https://devfeed.tech/tags/computer-vision.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [defense](<https://devfeed.tech/tags/defense.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [feature](<https://devfeed.tech/tags/feature.md>), [localization](<https://devfeed.tech/tags/localization.md>), [math-pr](<https://devfeed.tech/tags/math-pr.md>), [model](<https://devfeed.tech/tags/model.md>), [pixels](<https://devfeed.tech/tags/pixels.md>), [predictions](<https://devfeed.tech/tags/predictions.md>), [redirect](<https://devfeed.tech/tags/redirect.md>), [repair](<https://devfeed.tech/tags/repair.md>), [risk](<https://devfeed.tech/tags/risk.md>), [semantic](<https://devfeed.tech/tags/semantic.md>), [semantics](<https://devfeed.tech/tags/semantics.md>)

## AI overview

RePAIR is a post-hoc defense for backdoors in semantic segmentation that uses no labels and does not require knowing the attack. It infers the victim-to-target class relation from an unlabeled, potentially contaminated image pool, then identifies suspicious pixels by comparing classifier-head predictions with internal semantic evidence. Feature and output corrections steer those pixels toward inferred victim semantics; conservative localization and output constraints are intended to limit changes to benign predictions. The authors frame repair as an approximate reversal of adversarial probability transport and derive theoretical bounds and sufficient conditions. Experiments across segmentation architectures and attack settings report substantially reduced attack success while largely preserving clean performance.

## Source excerpt

arXiv:2610.37015v1 Announce Type: cross Abstract: Despite significant advances in computer vision, semantic segmentation models remain vulnerable to backdoor attacks, posing a security risk. These attacks redirect pixels of a chosen victim class to an attacker-specified target class when a trigger is present, while preserving clean predictions. Existing defenses are largely derived from image classification, while dedicated segmentation defenses remain scarce and typically require retraining or model modification. We introduce RePAIR, a post-hoc, label-free, attack-unaware defense based on the observation that backdoor-induced shifts can be stronger in the classifier head than in internal representations, leaving recoverable victim-class semantics. RePAIR infers the victim-target relation from an unlabeled, potentially contaminated image pool, constructs a safety-approved semantic reference, and localizes suspicious pixels by comparing head predictions with internal semantic evidence. It then applies feature-space and output corrections toward the inferred victim semantics, while conservative localization and output constraints limit changes to benign predictions. We model the attack as adversarial probability transport and repair as its approximate reversal, deriving bounds and sufficient conditions for localization, residual attack success, clean-side changes, defense-induced target predictions, and attack-unaware pair identification. Experiments across segmentation architectures and attack settings show that RePAIR substantially reduces attack success while largely preserving clean performance.