Research map / Adversarial robustness and security / Certified and empirical robustness
PixMix: Dreamlike Pictures Comprehensively Improve Safety Measures
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 PixMix dreamlike-picture augmentation | Wide ResNet-40-4 · CIFAR-100-C | mCE: 50.0 -> 30.5 | Helped |
| Add PixMix dreamlike-picture augmentation | Wide ResNet-40-4 · CIFAR-100-P | mFR: 10.7 -> 5.7 | Helped |
| Add PixMix dreamlike-picture augmentation | Wide ResNet-40-4 · CIFAR-100 PGD | Error: 96.8 -> 92.9 | Helped |
| Add PixMix dreamlike-picture augmentation | Wide ResNet-40-4 · CIFAR-100 | RMS calibration error: 31.2 -> 8.1 | Helped |
| Add PixMix dreamlike-picture augmentation | Wide ResNet-40-4 · CIFAR-100 OOD | AUROC: 77.7 -> 89.3 | Helped |
| Add PixMix dreamlike-picture augmentation | ResNet-50 · ImageNet-C | mCE: 78.2 -> 65.8 | Helped |
| Add PixMix dreamlike-picture augmentation | ResNet-50 · ImageNet | Clean error: 23.9 -> 22.6 | Helped |
| Use fractals and feature visualizations together instead of fractals alone (ablation) | PixMix · CIFAR-100-C | mCE: 32.3 -> 30.5 | Helped |
| Use fractals and feature visualizations together instead of feature visualizations alone (ablation) | PixMix · CIFAR-100-C | mCE: 30.3 -> 30.5 | No clear effect |
| Add mixing-set pictures to clean-image augmentations (ablation) | PixMix · CIFAR-100-C | mCE: 34.1 -> 30.5 | Helped |
| Also mix with augmented clean images instead of using mixing-set pictures alone (ablation) | PixMix · CIFAR-100-C | mCE: 31.1 -> 30.5 | Helped |
| Compute feature visualizations on the fly during training (ablation) | PixMix · CIFAR-100 OOD | AUROC: 89.3 -> 86.7 | Hurt |
Closest papers by meaning
- MultiOOD: Scaling Out-of-Distribution Detection for Multiple Modalities (NeurIPS 2024 · similarity 0.83)
- Spurious-Aware Prototype Refinement for Reliable Out-of-Distribution Detection (NeurIPS 2025 · similarity 0.82)
- RODEO: Robust Outlier Detection via Exposing Adaptive Out-of-Distribution Samples (ICML 2024 · similarity 0.82)
- ROME is Forged in Adversity: Robust Distilled Datasets via Information Bottleneck (ICML 2025 · similarity 0.82)
- A Noisy Elephant in the Room: Is Your Out-of-Distribution Detector Robust to Label Noise? (CVPR 2024 · similarity 0.82)
- Expecting The Unexpected: Towards Broad Out-Of-Distribution Detection (NeurIPS 2024 · similarity 0.82)
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.