Research map / Adversarial robustness and security / Black-box and transfer attacks
SegPAR: Class-Centric Decision-Based Sparse Attack for Semantic Segmentation
Dongsu Song, Daeyun Go, BOSEUNG SEO, Jay Hoon Jung
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 |
|---|---|---|---|
| shift from image-centric to class-centric exploration with class-wise inputs | RFPAR (DeepLabV3) · Cityscapes | R.MIoU / sparsity: R.MIoU 0.176 -> 0.101 (with R_d), sparsity 4.51% -> 3.33% | Improved |
| use discrepancy reward instead of standard reward in RFPAR (ablation) | RFPAR (DeepLabV3) · VOC2012 | R.MIoU: 0.466 -> 0.277 | Improved |
| use discrepancy reward instead of standard reward in RFPAR (ablation) | RFPAR (DeepLabV3) · Cityscapes | R.MIoU: 0.176 -> 0.210 | Hurt |
| use discrepancy reward instead of standard reward in Pixle (ablation) | Pixle (DeepLabV3) · VOC2012 | R.MIoU: 0.447 -> 0.279 | Improved |
| use discrepancy reward instead of standard reward in Pixle (ablation) | Pixle (DeepLabV3) · Cityscapes | R.MIoU: 0.394 -> 0.411 | Hurt |
| use discrepancy reward instead of standard reward in Sparse-RS (ablation) | Sparse-RS (DeepLabV3) · Cityscapes | R.MIoU: 0.172 -> 0.152 | Improved |
| use discrepancy reward instead of standard reward in Sparse-RS (ablation) | Sparse-RS (DeepLabV3) · VOC2012 | R.MIoU: 0.423 -> 0.258 | Improved |
| combine discrepancy reward with class-centric SegPAR on high-resolution images (ablation) | SegPAR · Cityscapes | R.MIoU: 0.210 (RFPAR+R_d) -> 0.101 (SegPAR+R_d) | Improved |
| apply SegPAR to adversarially trained models | SegPAR vs RFPAR · VOC2012 (DeepLabV3_DDCAT) | R.MIoU: 0.543 -> 0.297 | Improved |
| adversarial training against sparse attacks on low-resolution data | DeepLabV3 + SegPAR · VOC2012 | R.MIoU: 0.238 -> 0.297 (DDCAT) | No clear effect |
| adversarial training on high-resolution data | DeepLabV3 + SegPAR · Cityscapes | R.MIoU: 0.101 -> 0.428 (DDCAT) | Improved |
Closest papers by meaning
- Seg-ReSearch: Segmentation with Interleaved Reasoning and External Search (ICML 2026 · similarity 0.85)
- Attacks on Continual Semantic Segmentation by Perturbing Incremental Samples (AAAI 2024 · similarity 0.84)
- SED: A Simple Encoder-Decoder for Open-Vocabulary Semantic Segmentation (CVPR 2024 · similarity 0.84)
- Pixels to Tokens: Token Space Efficient Active Learning for Low-Budget Semantic Segmentation (NeurIPS 2026 · similarity 0.84)
- CosPGD: an efficient white-box adversarial attack for pixel-wise prediction tasks (ICML 2024 · similarity 0.84)
- Influencer Backdoor Attack on Semantic Segmentation (ICLR 2024 · similarity 0.83)
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.
