Research map / Adversarial robustness and security / Certified and empirical robustness
Amplitude-Phase Recombination: Rethinking Robustness of Convolutional Neural Networks in Frequency Domain
Guangyao Chen, Peixi Peng, Li Ma, Jia Li, Lin Du, Yonghong Tian
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 APR-S and APR-P together (APR-SP) to APR-S (ablation) | APR-S · CIFAR-100-C | mean corruption error: 36.8 -> 33.9 | Improved |
| add APR-S to APR-P (APR-SP) (ablation) | APR-P · CIFAR-10-C | mean corruption error: 19.7 -> 10.0 | Improved |
| replace Mixup with APR-P (amplitude swap, phase label) | Standard training · CIFAR-100-C | mean corruption error: 52.6 -> 46.6 | Improved |
| replace AugMix with APR-S | Standard training · CIFAR-100-C | mean corruption error: 38.3 -> 36.8 | Improved |
| add APR-SP on top of ImageNet augmentation | Standard ResNet-50 · ImageNet-C | mCE: 80.6 -> 65.0 | Improved |
| add APR-SP to DeepAugment | DeepAugment · ImageNet-C | mCE: 63.1 -> 57.5 | Improved |
| add APR-SP to DeepAugment | DeepAugment · ImageNet clean | test error: 26.3 -> 26.4 | No clear effect |
| add APR-SP to standard training | Standard ResNet-50 · ImageNet clean | test error: 23.9 -> 24.4 | Hurt |
| add APR-P instead of Cutout | CE · CIFAR-10 OOD detection | mean AUROC: 91.0 -> 93.1 | Improved |
| add APR-P to APR-S (APR-SP) (ablation) | APR-S · CIFAR-10 OOD detection | mean AUROC: 91.3 -> 94.7 | Improved |
| add APR-SP vs Cutout under FSGM adversarial training | FSGM · CIFAR-10 AutoAttack | robust accuracy: 41.6 -> 45.7 | Improved |
| add Cutout to FSGM | FSGM · CIFAR-10 AutoAttack | robust accuracy: 43.2 -> 41.6 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Mixup — mixup: Beyond Empirical Risk Minimization (ICLR 2018) · better on 2 of 2 shared results
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
- High-Frequency Component Helps Explain the Generalization of Convolutional Neural Networks (CVPR 2020 · similarity 0.91)
- HybridAugment++: Unified Frequency Spectra Perturbations for Model Robustness (ICCV 2023 · similarity 0.90)
- Domain Generalization with Vital Phase Augmentation (AAAI 2024 · similarity 0.85)
- D-GAP: Improving Out-of-Domain Robustness via Dataset-Agnostic and Gradient-Guided Augmentation in Amplitude and Pixel Spaces (NeurIPS 2026 · similarity 0.85)
- Diffusion-based Adversarial Purification from the Perspective of the Frequency Domain (ICML 2025 · similarity 0.85)
- Multiscale Low-Frequency Memory Network for Improved Feature Extraction in Convolutional Neural Networks (AAAI 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.
