Research map / Adversarial robustness and security / Attacks on text and graphs
Fast and Flexible Robustness Certificates for Semantic Segmentation
Thomas Massena, Corentin Friedrich, Franck Mamalet, Mathieu Serrurier
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 |
|---|---|---|---|
| Use Lipschitz bound certificate instead of randomized smoothing (SegCertify) | Lipschitz DeepLabV3 · Cityscapes eps=0.1 | Certification time: 62.1 s -> 0.1 s | Improved |
| Lipschitz bound vs SegCertify sigma=0.2 on CRPA | Lipschitz DeepLabV3 · Cityscapes eps=0.1 | CRPA: 83.13% (SegCertify) vs 81.80% (ours) | Mixed |
| Lipschitz bound vs SegCertify sigma=0.3 | Lipschitz DeepLabV3 · Cityscapes eps=0.1 | CRPA: 53.48% -> 81.80% | Improved |
| Use more than 1000 Monte Carlo samples for randomized smoothing on the same LipNet | SegCertify · Oxford-IIIT Pet | CRPA | Mixed |
| Replace unconstrained blocks with Lipschitz blocks (ablation) | DeepLabV3 · Cityscapes | Pixel accuracy: 94.41% -> 92.07% | Hurt |
| Replace unconstrained blocks with Lipschitz blocks (ablation) | DeepLabV3 · Cityscapes | mIoU: 64.55 -> 51.80 | Hurt |
| Use Lipschitz parametrization at batch size 30 vs 100 (ablation) | DeepLabV3 M1 · Oxford-IIIT Pet | Training overhead vs AllNet: 2.36x -> 1.09x | Mixed |
| Raise attack budget epsilon from 0 to 0.1 | Lipschitz DeepLabV3 · Cityscapes person class | Worst reachable class IoU: 47.37 -> 03.83 | Hurt |
| Use temperature-scaled cross-entropy training (ablation) | Lipschitz DeepLabV3 · Cityscapes | Certificate non-vacuity | Improved |
Closest papers by meaning
- Scaling Data-Driven Probabilistic Robustness Analysis for Semantic Segmentation Neural Networks (NeurIPS 2025 · similarity 0.89)
- Benchmarking the Robustness of Semantic Segmentation Models (CVPR 2020 · similarity 0.88)
- Evaluating the Adversarial Robustness of Semantic Segmentation: Trying Harder Pays Off (ECCV 2024 · similarity 0.88)
- Benchmarking the Robustness of Semantic Segmentation Models with Respect to Common Corruptions (IJCV 2020 · similarity 0.87)
- Reliability in Semantic Segmentation: Are We on the Right Track? (CVPR 2023 · similarity 0.87)
- Adaptive Hierarchical Certification for Segmentation using Randomized Smoothing (ICML 2024 · 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.
