Research map / Adversarial robustness and security / Certified and empirical robustness
StyleMix: Separating Content and Style for Enhanced Data Augmentation
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 class-distance-based automatic style-mixing control to StyleCutMix (ablation) | StyleCutMix · CIFAR-100 | Top-1 Err(%): 14.61 -> 14.17 | Helped |
| Add class-distance-based automatic style-mixing control to StyleCutMix (ablation) | StyleCutMix · CIFAR-10 | Top-1 Err(%): 2.79 -> 2.55 | Helped |
| Add class-distance-based automatic style-mixing control to StyleCutMix (ablation) | StyleCutMix · ImageNet | Top-1 Err(%): 23.42 -> 22.71 | Helped |
| Use content/style mixing with cut-and-paste rather than CutMix alone | PyramidNet-200 · CIFAR-100 | Top-1 Err(%): 14.47 -> 14.61 | Hurt |
| Use content/style mixing with cut-and-paste rather than CutMix alone | PyramidNet-200 · CIFAR-10 | Top-1 Err(%): 2.88 -> 2.79 | Helped |
| Use content/style mixing with cut-and-paste rather than CutMix alone | ResNet-50 · ImageNet | Top-1 Err(%): 22.92 -> 23.42 | Hurt |
| Add content/style mixing with automatic style control compared with CutMix | PyramidNet-200 · CIFAR-100 | Top-1 Err(%): 14.47 -> 14.17 | Helped |
| Add content/style mixing with automatic style control compared with CutMix | PyramidNet-200 · CIFAR-10 | Top-1 Err(%): 2.88 -> 2.55 | Helped |
| Add content/style mixing with automatic style control compared with Puzzle Mix | ResNet-50 · ImageNet | Top-1 Err(%): 22.49 -> 22.71 | Hurt |
| Add StyleCutMix augmentation instead of the vanilla baseline | PyramidNet-200 · CIFAR-100 FGSM (1), FGSM (2), and FGSM (4) | Top-1 Err(%): 63.68 -> 58.09; 74.70 -> 65.88; 80.39 -> 73.46 | Helped |
| Use automatic γ rather than randomly sampled γ in StyleCutMix (ablation) | PyramidNet-200 · CIFAR-100 FGSM (1), FGSM (2), and FGSM (4) | Top-1 Err(%): 58.09 -> 58.81; 65.88 -> 66.15; 73.46 -> 74.72 | Hurt |
| Vary δ for automatic style-mixing control (ablation) | StyleCutMix on CIFAR-100 · CIFAR-100 | Top-1 Err(%): δ = 1.5: 14.35; δ = 2.0: 14.52; δ = 3.0: 14.17; δ = 4.0: 14.42 | Mixed |
Closest papers by meaning
- Adversarial AutoMixup (ICLR 2024 · similarity 0.89)
- DiffuseMix: Label-Preserving Data Augmentation with Diffusion Models (CVPR 2024 · similarity 0.86)
- Enhance Image Classification via Inter-Class Image Mixup with Diffusion Model (CVPR 2024 · similarity 0.85)
- Catch-Up Mix: Catch-Up Class for Struggling Filters in CNN (AAAI 2024 · similarity 0.85)
- TdAttenMix: Top-Down Attention Guided Mixup (AAAI 2025 · similarity 0.85)
- CSGO: Content-Style Composition in Text-to-Image Generation (NeurIPS 2025 · similarity 0.84)
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.