Research map / Adversarial robustness and security / Attacks on text and graphs
Characterizing Adversarial Examples Based on Spatial Consistency Information for Semantic Segmentation
Chaowei Xiao, Ruizhi Deng, Bo Li, Fisher Yu, Mingyan Liu, Dawn Song
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| increase number of overlapping regions K from 1 to 5 (ablation) | spatial consistency detector · Cityscapes adversarial detection (DAG, Pure target, non-adaptive attack) | AUC (%): 91% -> 100% | Improved |
| increase K from 1 to 5 (ablation) | spatial consistency detector · Cityscapes adversarial detection (DAG, Pure target, non-adaptive attack) | AUC (%): 96% -> 100% | Improved |
| increase K beyond 5 to 50 (ablation) | spatial consistency detector · Cityscapes adversarial detection (DAG, Kitty target, non-adaptive attack) | AUC (%): 100% -> 100% | No clear effect |
| apply adaptive attack against scale-based detector (std 3.0) (ablation) | scale consistency detector · Cityscapes adversarial detection (DAG, Kitty target, adaptive attack) | AUC (%): 100% -> 0% | Hurt |
| apply adaptive attack against scale-based detector (std 0.5) (ablation) | scale consistency detector · Cityscapes adversarial detection (DAG, Kitty target, adaptive attack) | AUC (%): 95% -> 67% | Hurt |
| apply adaptive attack against spatial consistency detector (ablation) | spatial consistency detector · Cityscapes adversarial detection (DAG, Kitty target, adaptive attack) | AUC (%): 100% -> 100% | No clear effect |
| use larger Gaussian std for scale detector under adaptive attack (ablation) | scale consistency detector · Cityscapes adversarial detection (DAG, Kitty target, adaptive attack) | AUC (%): 67% -> 0% | Hurt |
| use random patch size 384-512 instead of fixed 512 (ablation) | spatial consistency detector · Cityscapes adversarial detection (DAG and Houdini, Kitty and Pure targets) | AUC (%) | No clear effect |
| replace spatial with scale-based detection (non-adaptive) (ablation) | detector · Cityscapes adversarial detection (DAG and Houdini, Kitty and Pure targets) | AUC (%) | No clear effect |
| increase K from 1 to 5 (ablation) | spatial consistency detector · Cityscapes adversarial detection (Houdini, Kitty target, adaptive attack) | AUC (%): 94% -> 100% | Improved |
Closest papers by meaning
- On the Robustness of Semantic Segmentation Models to Adversarial Attacks (CVPR 2018 · similarity 0.92)
- Adversarial Examples for Semantic Segmentation and Object Detection (ICCV 2017 · similarity 0.90)
- Indirect Local Attacks for Context-aware Semantic Segmentation Networks (ECCV 2020 · similarity 0.90)
- Evaluating the Adversarial Robustness of Semantic Segmentation: Trying Harder Pays Off (ECCV 2024 · similarity 0.89)
- RP-PGD: Boosting Segmentation Robustness with a Region-and-Prototype Based Adversarial Attack (AAAI 2025 · similarity 0.86)
- Constructing Semantics-Aware Adversarial Examples with a Probabilistic Perspective (NeurIPS 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
