Research map / Label-efficient and robust learning / Domain adaptation and generalization
OnlineAugment: Online Data Augmentation with Less Domain Knowledge
Zhiqiang Tang, Yunhe Gao, Leonid Karlinsky, Prasanna Sattigeri, Rogerio Feris, Dimitris Metaxas
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 |
|---|---|---|---|
| use double instead of single cycle-consistency in noise-conditioned A-STN (ablation) | A-STN · R-CIFAR-10 | accuracy (%): 82.62 -> 84.94 | Improved |
| use double instead of single cycle-consistency in image-conditioned A-STN (ablation) | A-STN · R-CIFAR-10 | accuracy (%): 83.24 -> 84.76 | Improved |
| condition A-STN on Gaussian noise instead of image (double cycle) (ablation) | A-STN · R-CIFAR-10 | accuracy (%): 84.76 -> 84.94 | No clear effect |
| add smoothness regularization to D-VAE (ablation) | D-VAE · R-CIFAR-10 | accuracy (%): 82.72 -> 82.86 | No clear effect |
| add smoothness regularization to D-VAE (ablation) | D-VAE · Liver segmentation | Dice (%): 68.51 -> 70.49 | Improved |
| use P-VAE noise generation instead of AdvProp with PGD | P-VAE · R-CIFAR-10 | accuracy (%): 83.00 -> 84.08 | Improved |
| add A-STN on top of AutoAugment | AutoAugment (Wide-ResNet-28-10) · R-CIFAR-10 | accuracy (%): 85.43 -> 89.39 | Improved |
| combine A-STN, D-VAE and P-VAE on top of AutoAugment | AutoAugment (Wide-ResNet-28-10) · R-CIFAR-100 | accuracy (%): 47.87 -> 53.72 | Improved |
| combine all three models on Shake-Shake CIFAR-100 over best single model | AutoAugment (Shake-Shake) · R-CIFAR-100 (Shake-Shake) | accuracy (%): 50.87 -> 50.11 | Hurt |
| add OnlineAugment on top of AutoAugment | AutoAugment · CIFAR-10 | top-1 error (%): 2.6 -> 2.0 | Improved |
| use OnlineAugment alone instead of AutoAugment | ResNet-50 · ImageNet | top-1 error (%): 22.4 -> 22.5 | No clear effect |
| combine A-STN and D-VAE instead of A-STN alone | OnlineAugment · LiTS (average Dice) | Dice (%): 72.13 -> 73.35 | Improved |
Closest papers by meaning
- AutoAugment: Learning Augmentation Strategies From Data (CVPR 2019 · similarity 0.90)
- Data-Efficient Augmentation for Training Neural Networks (NeurIPS 2022 · similarity 0.89)
- TeachAugment: Data Augmentation Optimization Using Teacher Knowledge (CVPR 2022 · similarity 0.89)
- Adversarial AutoAugment (ICLR 2020 · similarity 0.89)
- Faster AutoAugment: Learning Augmentation Strategies Using Backpropagation (ECCV 2020 · similarity 0.89)
- KeepAugment: A Simple Information-Preserving Data Augmentation Approach (CVPR 2021 · similarity 0.88)
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.
