What is Cutout / Random Erasing?
easyAnswer
- During training, mask a random rectangular region of the input image with zeros (Cutout) or random noise (Random Erasing).
- Forces the network to look at the whole image rather than one salient region — improves robustness to occlusion.
- Simple, cheap, effective.
- Standard in vision recipes since 2017; combines well with Mixup / CutMix.
Check yourself — multiple choice
- Removes the label
- Mask a random rectangular region with 0 or noise — forces robust global features and helps occlusion
- Same as Mixup
- Only used for text
Cutout / Random Erasing: mask a random rectangle → occlusion robustness.
#augmentation#regularization
Practise Deep Learning
214 interview questions in this topic.
Related questions
- How does Mixup work?
- Why is data augmentation effectively free regularization?
- What is consistency regularization?
- Your validation loss is lower than your training loss. Is something broken?
- How does dropout work and when is it applied?
- Why does AdamW work better than Adam + L2 weight decay for transformers?