Research map / Adversarial robustness and security / Certified and empirical robustness
PRIME: A Few Primitives Can Boost Robustness to Common Corruptions
Apostolos Modas, Rahul Rade, Guillermo Ortiz-Jiménez, Seyed-Mohsen Moosavi-Dezfooli, Pascal Frossard
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 |
|---|---|---|---|
| add spectral transformation to PRIME (ablation) | PRIME (ResNet-18) · ImageNet-100-C (no JSD) | corruption accuracy: 49.7 -> 64.1 | Improved |
| add spatial transformation alone (ablation) | PRIME (ResNet-18) · ImageNet-100-C (no JSD) | corruption accuracy: 49.7 -> 53.8 | Improved |
| add color transformation alone (ablation) | PRIME (ResNet-18) · ImageNet-100-C (no JSD) | corruption accuracy: 49.7 -> 59.9 | Improved |
| add color to spectral+spatial (ablation) | PRIME (ResNet-18) · ImageNet-100-C (no JSD) | corruption accuracy: 64.5 -> 68.8 | Improved |
| add spatial to spectral+color (ablation) | PRIME (ResNet-18) · ImageNet-100-C (no JSD) | corruption accuracy: 67.5 -> 68.8 | Improved |
| add primitives (all three) on clean accuracy (ablation) | PRIME (ResNet-18) · ImageNet-100 clean | accuracy: 88.0 -> 87.1 | Hurt |
| add mixing of transformed images (ablation) | PRIME · Min. cosine distance to ImageNet-C (SimCLRv2) | min cosine distance (x1e-3), lower better: 10.61 -> 7.71 | Improved |
| add mixing to AugMix (ablation) | AugMix · Min. cosine distance to ImageNet-C (SimCLRv2) | min cosine distance (x1e-3), lower better: 20.57 -> 17.48 | Improved |
| replace AugMix primitives with PRIME max-entropy primitives | AugMix-style mixing · Min. cosine distance to ImageNet-C (SimCLRv2) | min cosine distance (x1e-3), lower better: 17.48 -> 7.71 | Improved |
| replace AugMix with PRIME | ResNet-18 · ImageNet-100-C | corruption accuracy: 60.7 -> 71.6 | Improved |
| add PRIME on top of DeepAugment | ResNet-18 · ImageNet-100-C | mCE: 68.1 -> 54.6 | Improved |
| add BN-statistics adaptation with 8 corrupted samples | PRIME (ResNet-18) · ImageNet-100-C | corruption accuracy: 71.6 -> 75.3 | Improved |
Later papers that compared with this method
1 later accepted paper reports PRIME as a baseline in their results tables; 0 of them report a better number on most of the results they share with it.
- Feature Collapse Under Corruption: An Entropy Perspective on Robust Neural Networks (ICML 2026 · better on 0 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Improving robustness to corruptions with multiplicative weight perturbations (NeurIPS 2024 · similarity 0.88)
- On Interaction Between Augmentations and Corruptions in Natural Corruption Robustness (NeurIPS 2021 · similarity 0.88)
- Dynamic Sparse Training versus Dense Training: The Unexpected Winner in Image Corruption Robustness (ICLR 2025 · similarity 0.87)
- Slight Corruption in Pre-training Data Makes Better Diffusion Models (NeurIPS 2024 · similarity 0.85)
- Adaptive Sensitivity Analysis for Robust Augmentation against Natural Corruptions in Image Segmentation (ICML 2025 · similarity 0.85)
- Adaptive Sharpness-Aware Pruning for Robust Sparse Networks (ICLR 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.
