Research map / Adversarial robustness and security / Certified and empirical robustness
HybridAugment++: Unified Frequency Spectra Perturbations for Model Robustness
Mehmet Kerim Yucel, Ramazan Gokberk Cinbis, Pinar Duygulu
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 HybridAugment++ (phase/amplitude swap on low-frequency component) on top of paired and single-image augmentation (ablation) | HybridAugment · CIFAR-10-C (mean over 5 architectures) | mCE: 9.2 -> 8.9 | Improved |
| lift same-class restriction in hybrid image sampling versus RFC (paired-only) (ablation) | HybridAugment · CIFAR-10-C (mean over 5 architectures) | mCE: 20.3 -> 14.5 | Improved |
| add HybridAugment++ to ResNet50 on ImageNet | ResNet50 · ImageNet-C | mCE: 80.6 -> 67.3 | Improved |
| combine HA++_PS with DeepAugment and AugMix | HybridAugment++ · ImageNet-C | mCE: 62.0 -> 56.4 | Improved |
| use higher cut-off frequency (stronger blur) in HA++_PS on ImageNet (ablation) | HybridAugment++ · ImageNet-C | mCE: 67.3 -> 65.8 (clean error 23.7 -> 25.5) | Mixed |
| use higher cut-off frequency on ImageNet-C-bar (ablation) | HybridAugment++ · ImageNet-C-bar | error: 52.2 -> 53.4 | Hurt |
| use kernel size K=3 with sigma 0.5 in Gaussian low-pass | HybridAugment++ · CIFAR-10/100 (ResNet18) | clean accuracy and mCE | Improved |
| use high-frequency image labels instead of low-frequency image labels | HybridAugment · CIFAR | overall performance | Hurt |
| add HybridAugment++ to adversarial training on CIFAR-10 | FGSM adversarial training · CIFAR-10 AutoAttack | robust accuracy: 43.2 -> 46.0 | Improved |
| add HA++_PS to adversarial training versus HA_S | HybridAugment++ · CIFAR-10 AutoAttack | clean accuracy: 86.5 -> 82.8 | Hurt |
| train with HybridAugment variants for OOD detection | ResNet18 (CIFAR-10) · OOD detection (mean AUROC) | AUROC: 88.1 -> 94.7 | Improved |
| paired-only HybridAugment variants on OOD detection | HybridAugment · OOD detection (mean AUROC) | AUROC: 88.1 -> 89.0 | No clear effect |
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.
- PixMix — PixMix: Dreamlike Pictures Comprehensively Improve Safety Measures (CVPR 2022) · better on 1 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
- D-GAP: Improving Out-of-Domain Robustness via Dataset-Agnostic and Gradient-Guided Augmentation in Amplitude and Pixel Spaces (NeurIPS 2026 · similarity 0.87)
- Fourier-basis Functions to Bridge Augmentation Gap: Rethinking Frequency Augmentation in Image Classification (CVPR 2024 · similarity 0.87)
- Frequency-Aware Deepfake Detection: Improving Generalizability through Frequency Space Domain Learning (AAAI 2024 · similarity 0.86)
- Multiscale Low-Frequency Memory Network for Improved Feature Extraction in Convolutional Neural Networks (AAAI 2024 · similarity 0.86)
- Do ImageNet-trained Models Learn Shortcuts? The Impact of Frequency Shortcuts on Generalization (CVPR 2025 · similarity 0.86)
- Frequency-Adaptive Dilated Convolution for Semantic Segmentation (CVPR 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.
