Research map / Label-efficient and robust learning / Recognition and long-tail
Saliency Grafting: Innocuous Attribution-Guided Mixup with Calibrated Label Mixing
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 |
|---|---|---|---|
| Add ShakeDrop regularization to Saliency Grafting | PyramidNet-200 on CIFAR-100 · CIFAR-100 | Top-1 error (%): 13.94 -> 13.05 | Helped |
| Use stochastic rather than deterministic region selection (ablation) | Saliency Grafting with area-based labels on CIFAR-100 · CIFAR-100 | Top-1 error (%): 16.34 -> 15.67; 14.63 -> 14.14 | Helped |
| Use saliency-based rather than area-based label mixing (ablation) | Saliency Grafting with stochastic selection on CIFAR-100 · CIFAR-100 | Top-1 error (%): 15.67 -> 15.32; 14.14 -> 13.94 | Helped |
| Replace forward saliency maps with backward CAM maps (ablation) | Saliency Grafting on CIFAR-100 · CIFAR-100 | Top-1 error (%): 15.32 -> 15.70 | Hurt |
| Lower the saliency threshold below the normalized saliency mean (ablation) | Saliency Grafting on CIFAR-100 · CIFAR-100 | Top-1 error (%) | Hurt |
| Set the saliency threshold to zero, making grafting saliency-agnostic (ablation) | Saliency Grafting on CIFAR-100 · CIFAR-100 | Top-1 error (%) | Hurt |
| Increase the number of independently augmented batches (ablation) | Saliency Grafting on 20% of CIFAR-100 · CIFAR-100, 20% data | Top-1 error (%) | Helped |
| Increase the number of independently augmented batches (ablation) | PuzzleMix on 20% of CIFAR-100 · CIFAR-100, 20% data | Top-1 error (%) | No clear effect |
| Replace the image-resizing policy with fixed-size images (ablation) | Saliency Grafting on ImageNet · ImageNet | Top-1 error (%): 22.35 -> 22.26 | Helped |
| Replace the image-resizing policy with fixed-size images (ablation) | Saliency Grafting on ImageNet · ImageNet | Top-5 error (%): 6.19 -> 6.29 | Hurt |
| Increase the softmax temperature from 0.01 toward higher values (ablation) | Saliency Grafting on CIFAR-100 with ResNet-18 · CIFAR-100 | Top-1 error (%) | Helped |
Closest papers by meaning
- SaliencyMix: A Saliency Guided Data Augmentation Strategy for Better Regularization (ICLR 2021 · similarity 0.89)
- Puzzle Mix: Exploiting Saliency and Local Statistics for Optimal Mixup (ICML 2020 · similarity 0.88)
- GuidedMixup: An Efficient Mixup Strategy Guided by Saliency Maps (AAAI 2023 · similarity 0.88)
- Catch-Up Mix: Catch-Up Class for Struggling Filters in CNN (AAAI 2024 · similarity 0.87)
- ADD: Attribution-Driven Data Augmentation Framework for Boosting Image Super-Resolution (CVPR 2025 · similarity 0.85)
- Adversarial AutoMixup (ICLR 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.