Research map / Adversarial robustness and security / Attacks on text and graphs
Adversarial Attack and Disturbance Detection by Hadamard-Coded Output Representations for Object Detection and Semantic Segmentation
Lucas Görnhardt, Timo Bartels, Niklas Schwarz, Tim Fingscheidt
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 |
|---|---|---|---|
| replace one-hot output with Hadamard code and quantile detector | SegFormer MiT-B0 · Cityscapes val, all perturbations | Global AuROC: 72.1 -> 80.2 | Improved |
| replace one-hot output with Hadamard code and error detector | DeepLabv3+ · Cityscapes val, all perturbations | Global AuROC: 72.5 -> 80.4 | Improved |
| replace one-hot output with Hadamard code and error detector | SegFormer MiT-B0 · BDD100K val, all perturbations | Global AuROC: 62.0 -> 70.3 | Improved |
| replace one-hot output with Hadamard code and error detector | DeepLabv3+ · BDD100K val, all perturbations | Global AuROC: 59.2 -> 70.9 | Improved |
| replace one-hot output with Hadamard code | SegFormer MiT-B0 · Cityscapes val, clean | mIoU: 76.4 -> 76.5 | No clear effect |
| replace one-hot output with Hadamard code | DeepLabv3+ · Cityscapes val, clean | mIoU: 80.2 -> 80.1 | No clear effect |
| replace one-hot output with Hadamard code | SegFormer MiT-B0 · BDD100K val, clean | mIoU: 59.5 -> 57.6 | Hurt |
| replace one-hot output with Hadamard code | DeepLabv3+ · BDD100K val, clean | mIoU: 59.8 -> 57.2 | Hurt |
| use quantile anomaly score instead of regression score (ablation) | HadamardNet SegFormer MiT-B0 · Cityscapes val, all perturbations | Global AuROC: 74.6 -> 80.2 | Improved |
| use quantile anomaly score instead of error score (ablation) | HadamardNet SegFormer MiT-B0 · Cityscapes val, Metzen attack | Global AuROC: 68.4 -> 75.9 | Improved |
| use Hadamard quantile detector instead of entropy | SegFormer MiT-B0 · Cityscapes val, Metzen attack | Global AuROC: 45.0 -> 75.9 | Improved |
| use Hadamard quantile detector instead of FGSM-optimized detector of Maag et al. | SegFormer MiT-B0 · Cityscapes val, FGSM attack | Global AuROC: 98.2 -> 95.9 | Hurt |
Closest papers by meaning
- Evaluating the Adversarial Robustness of Semantic Segmentation: Trying Harder Pays Off (ECCV 2024 · similarity 0.88)
- Adversarial Examples for Semantic Segmentation and Object Detection (ICCV 2017 · similarity 0.87)
- Towards Reliable Evaluation and Fast Training of Robust Semantic Segmentation Models (ECCV 2024 · similarity 0.86)
- Membership Inference Attacks against Large Vision-Language Models (NeurIPS 2024 · similarity 0.85)
- ASAM: Boosting Segment Anything Model with Adversarial Tuning (CVPR 2024 · similarity 0.85)
- Chain of Attack: On the Robustness of Vision-Language Models Against Transfer-Based Adversarial Attacks (CVPR 2025 · similarity 0.85)
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.
