Research map / Detection and segmentation / CNN features
RandAugment: Practical Automated Data Augmentation with a Reduced Search Space
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 |
|---|---|---|---|
| Use RandAugment instead of default augmentation | EfficientNet-B7 · ImageNet | top-1 accuracy: 84.0 -> 85.0 | Helped |
| Use RandAugment instead of AutoAugment | EfficientNet-B7 · ImageNet | top-1 accuracy: 84.4 -> 85.0 | Helped |
| Use RandAugment instead of AutoAugment | RetinaNet ResNet-101 · COCO | mAP: 40.4 -> 40.1 | Hurt |
| Use RandAugment instead of AutoAugment | RetinaNet ResNet-200 · COCO | mAP: 42.1 -> 41.9 | Hurt |
| Increase model size | Wide-ResNet on CIFAR-10 · CIFAR-10 | optimal distortion magnitude | Helped |
| Increase training-set size | Wide-ResNet-28-10 · CIFAR-10 | optimal distortion magnitude | Helped |
| Add rotate to a random subset of transformations (ablation) | RandAugment, Wide-ResNet-28-2 · CIFAR-10 | validation accuracy improvement: +1.3% | Helped |
| Add posterize to a random subset of transformations (ablation) | RandAugment, Wide-ResNet-28-2 · CIFAR-10 | validation accuracy improvement: -0.3% | Hurt |
| Add contrast transformation to a random subset (ablation) | RandAugment, Wide-ResNet-28-2 · CIFAR-10 | validation accuracy improvement: 0.0% | No clear effect |
| Use constant magnitude instead of random magnitude (ablation) | RandAugment, Wide-ResNet-28-10 · CIFAR-10 | validation accuracy: 97.3 -> 97.2 | No clear effect |
| Use linearly increasing magnitude instead of random magnitude (ablation) | RandAugment, Wide-ResNet-28-10 · CIFAR-10 | validation accuracy: 97.3 -> 97.2 | No clear effect |
| Learn transformation probabilities with first-order density matching instead of uniform probabilities (ablation) | RandAugment, Wide-ResNet-28-2 · CIFAR-10 | test accuracy: 95.8 -> 96.1 | Helped |
Closest papers by meaning
- AutoAugment: Learning Augmentation Strategies From Data (CVPR 2019 · similarity 0.92)
- Scale-Aware Automatic Augmentation for Object Detection (CVPR 2021 · similarity 0.89)
- MetaAugment: Sample-Aware Data Augmentation Policy Learning (AAAI 2021 · similarity 0.88)
- Tied-Augment: Controlling Representation Similarity Improves Data Augmentation (ICML 2023 · similarity 0.87)
- Your Image is My Video: Reshaping the Receptive Field via Image-To-Video Differentiable AutoAugmentation and Fusion (CVPR 2024 · similarity 0.85)
- SELECT: A Large-Scale Benchmark of Data Curation Strategies for Image Classification (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.