Research map / Detection and segmentation / CNN features
Random Erasing Data Augmentation
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 Random Erasing | WRN-28-10 · CIFAR-10 | top-1 error (%): 3.80 $\pm$ 0.07 -> 3.08 $\pm$ 0.05 | Helped |
| Add Random Erasing | WRN-28-10 · CIFAR-100 | top-1 error (%): 18.49 $\pm$ 0.11 -> 17.73 $\pm$ 0.15 | Helped |
| Add Random Erasing | WRN-28-10 · Fashion-MNIST | top-1 error (%): 4.01 $\pm$ 0.10 -> 3.65 $\pm$ 0.03 | Helped |
| Use ImageNet mean values rather than random values in erased pixels (ablation) | Random Erasing on ResNet-18-PreAct · CIFAR-10 | top-1 error (%): 4.31 $\pm$ 0.07 -> 4.35 $\pm$ 0.12 | No clear effect |
| Fill erased pixels with 0 rather than random values (ablation) | Random Erasing on ResNet-18-PreAct · CIFAR-10 | top-1 error (%): 4.31 $\pm$ 0.07 -> 4.62 $\pm$ 0.09 | Hurt |
| Fill erased pixels with 255 rather than random values (ablation) | Random Erasing on ResNet-18-PreAct · CIFAR-10 | top-1 error (%): 4.31 $\pm$ 0.07 -> 4.85 $\pm$ 0.13 | Hurt |
| Apply dropout on the image layer instead of rectangular erasing (ablation) | ResNet-18-PreAct · CIFAR-10 | top-1 error (%): 5.17 $\pm$ 0.18 -> 5.37 $\pm$ 0.12 | Hurt |
| Add random pixel noise instead of rectangular erasing (ablation) | ResNet-18-PreAct · CIFAR-10 | top-1 error (%): 5.17\% -> 6.52\% | Hurt |
| Combine Random Erasing with random flipping and random cropping (ablation) | ResNet-18-PreAct · CIFAR-10 | top-1 error (%): 5.17 $\pm$ 0.18 -> 4.31 $\pm$ 0.07 | Helped |
| Train with Random Erasing | ResNet-18-PreAct · CIFAR-10 with occluded area l = 0.5 | error rate (%): 75.04\% -> 56.36\% | Helped |
| Use object-box-aware erasing rather than whole-image-only erasing (ablation) | Fast R-CNN, VGG16 · VOC 2007 test | mAP (%): 70.5 -> 71.0 | Helped |
| Apply erasing to both whole images and object boxes (ablation) | Fast R-CNN, VGG16 · VOC 2007 test | mAP (%): 71.0 -> 71.5 | Helped |
Closest papers by meaning
- You Only Erase Once: Erasing Anything without Bringing Unexpected Content (CVPR 2026 · similarity 0.83)
- Learning by Erasing: Conditional Entropy Based Transferable Out-of-Distribution Detection (AAAI 2024 · similarity 0.83)
- Attentive Eraser: Unleashing Diffusion Model’s Object Removal Potential via Self-Attention Redirection Guidance (AAAI 2025 · similarity 0.81)
- All but One: Surgical Concept Erasing with Model Preservation in Text-to-Image Diffusion Models (AAAI 2024 · similarity 0.81)
- ESC: Erasing Space Concept for Knowledge Deletion (CVPR 2025 · similarity 0.80)
- Spurious Feature Eraser: Stabilizing Test-Time Adaptation for Vision-Language Foundation Model (AAAI 2025 · similarity 0.80)
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.