RePAIR: Inverting Backdoor Attacks in Semantic Segmentation

Summary

RePAIR is a post-hoc, label-free defense against backdoor attacks on semantic segmentation models, which redirect pixels of a victim class to an attacker-chosen target class when a trigger is present. It infers the victim-target relation from an unlabeled image pool, then localizes suspicious pixels and applies feature-space and output corrections toward the inferred victim semantics. Experiments across segmentation architectures and attack settings show it substantially reduces attack success while largely preserving clean performance.

Mitigation

RePAIR is the defense proposed by the source, a post-hoc, label-free, attack-unaware method that requires no retraining or model modification. The source does not describe a separate fix or configuration step.

Topics

• Adversarial machine learning

Related items

• HighCVE-2026-118378: Tabularis SQLite driver SQL injection through MCP describe_table toolSimilar attack · NVD/CVE Database
• HighGHSA-m65p-9j7q-6w33: MLflow vulnerable to arbitrary file read on unauthenticated servers via prompt model version sourceSimilar attack · GitHub Advisory Database
• CriticalGHSA-26p8-2jq9-3vq9: MLflow vulnerable to remote code execution on unauthenticated servers via third-party scorer deserialization importing caller-placed artifact modulesSimilar attack · GitHub Advisory Database
• MediumGHSA-w4fh-qvv9-3v23: Model Context Protocol Python SDK: Server bearer-token authentication accepts tokens issued for a different resource serverSimilar attack · GitHub Advisory Database
• MediumGHSA-5h93-6whr-6q8j: Model Context Protocol Python SDK: HTTP client transports followed cross-origin redirects carrying custom headers and bodiesSimilar attack · GitHub Advisory Database