Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Poisoning the Pixels: Revisiting Backdoor Attacks on Semantic Segmentation

Guangsheng Zhang, Huan Tian, Leo Zhang, Tianqing Zhu, Ming Ding, Wanlei Zhou, Bo Liu

Abstract

Semantic segmentation models are widely deployed in safety-critical applications such as autonomous driving, yet their vulnerability to backdoor attacks remains largely underexplored. Prior segmentation backdoor studies transfer threat settings from existing image classification tasks, focusing primarily on object-to-background mis-segmentation. In this work, we revisit the threats by systematically examining backdoor attacks tailored to semantic segmentation. We identify four coarse-grained attack vectors (Object-to-Object, Object-to-Background, Background-to-Object, and Background-to-Background attacks), as well as two fine-grained vectors (Instance-Level and Conditional attacks). To formalize these attacks, we introduce BADSEG, a unified framework that optimizes trigger designs and applies label manipulation strategies to maximize attack performance while preserving victim model utility. Extensive experiments across diverse segmentation architectures on benchmark datasets demonstrate that BADSEG achieves high attack effectiveness with minimal impact on clean samples. We further evaluate six representative defenses and find that they fail to reliably mitigate our attacks, revealing critical gaps in current defenses. Finally, we demonstrate that these vulnerabilities persist in recent emerging architectures, including transformer-based networks and the Segment Anything Model (SAM), thereby compromising their security. Our work reveals previously overlooked security vulnerabilities in semantic segmentation, and motivates the development of defenses tailored to segmentation-specific threat models.

Poisoning the Pixels: Revisiting Backdoor Attacks on Semantic Segmentation

Abstract

Semantic segmentation models are widely deployed in safety-critical applications such as autonomous driving, yet their vulnerability to backdoor attacks remains largely underexplored. Prior segmentation backdoor studies transfer threat settings from existing image classification tasks, focusing primarily on object-to-background mis-segmentation. In this work, we revisit the threats by systematically examining backdoor attacks tailored to semantic segmentation. We identify four coarse-grained attack vectors (Object-to-Object, Object-to-Background, Background-to-Object, and Background-to-Background attacks), as well as two fine-grained vectors (Instance-Level and Conditional attacks). To formalize these attacks, we introduce BADSEG, a unified framework that optimizes trigger designs and applies label manipulation strategies to maximize attack performance while preserving victim model utility. Extensive experiments across diverse segmentation architectures on benchmark datasets demonstrate that BADSEG achieves high attack effectiveness with minimal impact on clean samples. We further evaluate six representative defenses and find that they fail to reliably mitigate our attacks, revealing critical gaps in current defenses. Finally, we demonstrate that these vulnerabilities persist in recent emerging architectures, including transformer-based networks and the Segment Anything Model (SAM), thereby compromising their security. Our work reveals previously overlooked security vulnerabilities in semantic segmentation, and motivates the development of defenses tailored to segmentation-specific threat models.
Paper Structure (31 sections, 6 equations, 14 figures, 25 tables)

This paper contains 31 sections, 6 equations, 14 figures, 25 tables.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Semantic Segmentation
  4. Backdoor Attacks on Segmentation
  5. Threat Model
  6. Attack Vectors
  7. Coarse-Grained Backdoor Attacks
  8. Fine-Grained Backdoor Attacks
  9. BADSEG
  10. Stage 1: Backdoor Preparation
  11. Stage 2: Trigger Optimization
  12. Stage 3: Label Manipulation
  13. Backdoor Implementation
  14. Evaluation
  15. Experimental Setup
  16. ...and 16 more sections

Figures (14)

  • Figure 1: Illustration of coarse-grained and fine-grained attacks. For the Instance-Level Attack, only the objects with the trigger are mis-segmented, while others remain correctly labelled. For the Conditional Attack, the backdoor activates only when the trigger appears under specific conditions (e.g., red cars).
  • Figure 2: Overall workflow of the proposed BADSEG.
  • Figure 3: Normalized distance matrix across class pairs. Smaller values indicate stronger semantic correlations.
  • Figure 4: Euclidean distance of victim--target pairs and ASRs for O2B Attacks in \ref{['tab:poisoning_o2b_target_unchanged_sub']}.
  • Figure 5: KDE plots of training losses for ABL, entropy scores for STRIP, and detection scores for TeCo.
  • ...and 9 more figures