Research map / Label-efficient and robust learning / Recognition and long-tail
TrivialAugment: Tuning-Free Yet State-of-the-Art Data Augmentation
Samuel G. Müller, Frank Hutter
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 |
|---|---|---|---|
| use TA instead of brute-force RA search (91 trainings) | Wide-ResNet-40-2 · CIFAR-10 | accuracy: 96.42 -> 96.62 | Improved |
| use Wide augmentation space instead of RA space (ablation) | TA (Wide-ResNet-28-10) · CIFAR-100 | accuracy: 83.54 -> 84.33 | Improved |
| restrict to Full augmentation space (adds flips, blur, smooth) instead of RA space (ablation) | TA (Wide-ResNet-28-10) · SVHN Core | accuracy: 98.05 -> 97.63 | Hurt |
| use Full augmentation space instead of RA space (ablation) | TA (Wide-ResNet-28-10) · CIFAR-10 | accuracy: 97.46 -> 97.24 | Hurt |
| remove invert from AA space (ablation) | TA (Wide-ResNet-28-10) · CIFAR-10 | accuracy: 97.47 -> 97.55 | Improved |
| remove invert from AA space (ablation) | TA (Wide-ResNet-28-10) · SVHN Core | accuracy: 98.04 -> 97.97 | Hurt |
| sample only strength 30 instead of 0-30 (ablation) | TA (RA) · CIFAR-100 | accuracy: 83.54 -> 82.98 | Hurt |
| sample only strength 30 instead of 0-30 (ablation) | TA (RA) · SVHN Core | accuracy: 98.05 -> 98.16 | Improved |
| sample only strength 30 instead of 0-30 (ablation) | TA (RA) · CIFAR-10 | accuracy: 97.46 -> 97.45 | No clear effect |
| reduce strengths to {0,30} from 31 values (ablation) | TA (RA) · CIFAR-100 | accuracy: 83.54 -> 83.46 | No clear effect |
| use TA x8 batch augmentation instead of Adv. AA | Wide-ResNet-28-10 · CIFAR-100 | accuracy: 84.51 -> 84.62 | No clear effect |
| use TA instead of brute-force RA | Wide-ResNet-28-10 · Occluded CIFAR-10 | accuracy: 95.52 -> 95.72 | Improved |
Closest papers by meaning
- AutoAugment: Learning Augmentation Strategies From Data (CVPR 2019 · similarity 0.91)
- RandAugment: Practical Automated Data Augmentation with a Reduced Search Space (NeurIPS 2020 · similarity 0.90)
- FreeAugment: Data Augmentation Search Across All Degrees of Freedom (ECCV 2024 · similarity 0.88)
- Scale-Aware Automatic Augmentations for Object Detection With Dynamic Training (TPAMI 2022 · similarity 0.88)
- PseudoAugment: Learning to Use Unlabeled Data for Data Augmentation in Point Clouds (ECCV 2022 · similarity 0.88)
- Faster AutoAugment: Learning Augmentation Strategies Using Backpropagation (ECCV 2020 · similarity 0.87)
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.
