Research map / Detection and segmentation / CNN features
SaliencyMix: A Saliency Guided Data Augmentation Strategy for Better Regularization
Read the paperCheck 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 |
|---|---|---|---|
| Replace random source-patch selection with a patch centered on the source image's most salient region (ablation) | CutMix-style augmentation · CIFAR-10 and Tiny-ImageNet classification | classification error | Helped |
| Select source patches from non-salient regions instead of salient regions (ablation) | SaliencyMix · CIFAR-10 and Tiny-ImageNet classification | classification error | Hurt |
| Mix the salient source patch into the target's salient region rather than the corresponding location (ablation) | SaliencyMix · CIFAR-10 and Tiny-ImageNet classification | classification error | Hurt |
| Mix the salient source patch into a non-salient target region rather than the corresponding location (ablation) | SaliencyMix · CIFAR-10 and Tiny-ImageNet classification | classification error | Hurt |
| Use salient-to-corresponding patch placement instead of the other tested placement schemes (ablation) | SaliencyMix · CIFAR-10 and Tiny-ImageNet classification | classification error | Helped |
| Use a saliency detection algorithm that performs best in the reported comparison (ablation) | SaliencyMix · CIFAR-10 and Tiny-ImageNet classification | classification error | Helped |
| Use SaliencyMix rather than CutMix | ResNet-18 · CIFAR-10 | top-1 error (%): 9.44$\pm$0.34 -> 7.59$\pm$0.22 | Helped |
| Use SaliencyMix rather than CutMix | WideResNet-28-10 · CIFAR-10+ | top-1 error (%): 2.87$\pm$0.16 -> 2.76$\pm$0.07 | Mixed |
| Use SaliencyMix rather than CutMix | WideResNet-28-10 · CIFAR-100 | top-1 error (%): 23.21$\pm$0.20 -> 19.45$\pm$0.32 | Helped |
| Use SaliencyMix rather than PuzzleMix | WideResNet-28-10 · CIFAR-100+ | top-1 error (%): 16.23 -> 16.56$\pm$0.17 | Hurt |
| Use SaliencyMix rather than CutMix | ResNet-50 · ImageNet | top-1 error (%): 21.40 -> 21.26 | Helped |
| Use SaliencyMix rather than CutMix | ResNet-50 · ImageNet validation, FGSM adversarial examples | top-1 accuracy (%): 31.0 -> 32.96 | Helped |
Closest papers by meaning
- Saliency Grafting: Innocuous Attribution-Guided Mixup with Calibrated Label Mixing (AAAI 2022 · similarity 0.89)
- GuidedMixup: An Efficient Mixup Strategy Guided by Saliency Maps (AAAI 2023 · similarity 0.88)
- ADD: Attribution-Driven Data Augmentation Framework for Boosting Image Super-Resolution (CVPR 2025 · similarity 0.86)
- Catch-Up Mix: Catch-Up Class for Struggling Filters in CNN (AAAI 2024 · similarity 0.85)
- DiffuseMix: Label-Preserving Data Augmentation with Diffusion Models (CVPR 2024 · similarity 0.85)
- Saliuitl: Ensemble Salience Guided Recovery of Adversarial Patches against CNNs (CVPR 2025 · 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.