Research map / Label-efficient and robust learning / Recognition and long-tail
Regularizing Deep Networks with Semantic Data Augmentation
Yulin Wang, Gao Ming Huang, Shiji Song, Xuran Pan, Yitong Xia, Cheng Wu
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 ISDA to training | ResNet-50 · ImageNet val | top-1 error (%): 23.0 -> 21.9 | Improved |
| add ISDA to training | DenseNet-BC-121 · ImageNet val | top-1 error (%): 23.7 -> 23.2 | Improved |
| add ISDA to AutoAugment | Shake-Shake (26, 2x112d) · CIFAR-100 | test error (%): 15.21 -> 13.87 | Improved |
| add ISDA to VAT | CNN-13 · CIFAR-10 (1,000 labels) | test error (%): 18.12 -> 14.38 | Improved |
| add ISDA to PSPNet | PSPNet ResNet-101 · Cityscapes val | mIoU (%): 77.46 -> 78.72 | Improved |
| use dynamically estimated per-class covariance instead of identity matrix (ablation) | ISDA · CIFAR-100 | test error (%): 18.53 -> 17.98 | Improved |
| use full covariance instead of diagonal matrix (ablation) | ISDA · CIFAR-100 + Cutout | test error (%): 17.54 -> 16.95 | Improved |
| use per-class covariance instead of a single global covariance (ablation) | ISDA · CIFAR-100 + Cutout | test error (%): 18.12 -> 16.95 | Improved |
| use iteration-dependent lambda_0 instead of constant lambda_0 (ablation) | ISDA · CIFAR-100 + Cutout | test error (%): 17.34 -> 16.95 | Improved |
| increase lambda_0 to 10 (ablation) | ISDA · CIFAR-100 sensitivity | test error | Hurt |
| use ISDA upper bound (M to infinity) instead of sampled explicit augmentation with small M (ablation) | explicit SDA · CIFAR-100 + Cutout / AutoAugment | test error | Improved |
| use ISDA instead of GAN-based augmentation | ResNet-110 · CIFAR-100 | test error (%): 27.57 -> 28.25 | Improved |
Closest papers by meaning
- Implicit Semantic Data Augmentation for Deep Networks (NeurIPS 2019 · similarity 0.96)
- MetaSAug: Meta Semantic Augmentation for Long-Tailed Visual Recognition (CVPR 2021 · similarity 0.89)
- Data-Efficient Augmentation for Training Neural Networks (NeurIPS 2022 · similarity 0.89)
- Effective Data Augmentation With Diffusion Models (ICLR 2024 · similarity 0.86)
- SF(DA)$^2$: Source-free Domain Adaptation Through the Lens of Data Augmentation (ICLR 2024 · similarity 0.86)
- LLM-AutoDA: Large Language Model-Driven Automatic Data Augmentation for Long-tailed Problems (NeurIPS 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.
