Random erasing in Computer Vision - Cheat Sheet & Quick Revision

Recall & Review

beginner

What is Random Erasing in computer vision?

Random Erasing is a data augmentation technique where a random rectangle area in an image is erased (filled with random values or a constant) during training to help the model learn better and avoid overfitting.

beginner

Why do we use Random Erasing during training?

We use Random Erasing to make the model more robust by forcing it to focus on different parts of the image, not just the most obvious features. This helps the model generalize better to new images.

intermediate

How does Random Erasing differ from other augmentations like flipping or rotation?

Unlike flipping or rotation which change the whole image orientation, Random Erasing removes or hides a small part of the image randomly, simulating occlusion or missing parts to improve model resilience.

intermediate

What parameters control the Random Erasing process?

Key parameters include the size range of the erased area, the aspect ratio of the erased rectangle, and the probability of applying erasing to each image during training.

beginner

Can Random Erasing be used during model testing or only training? Why?

Random Erasing is used only during training to improve learning. During testing, images should remain unchanged to fairly evaluate the model's performance on real data.

What is the main goal of Random Erasing in training?

A To flip the image horizontally

B To increase image brightness

C To hide parts of the image randomly to improve model robustness

D To crop the image to a smaller size

Random Erasing hides random parts of the image to help the model learn to recognize objects even if parts are missing.

Which of these is NOT a parameter of Random Erasing?

A Probability of erasing

B Number of image channels

C Color of erased area

D Size of erased area

Number of image channels is a property of the image, not a parameter controlling Random Erasing.

When should Random Erasing be applied?

A Only during training

B Only during testing

C During both training and testing

D Never

Random Erasing is a training augmentation technique and should not be applied during testing.

Random Erasing helps the model to:

A Increase image resolution

B Ignore all image features

C Focus only on the center of the image

D Generalize better by learning from incomplete images

By erasing parts randomly, the model learns to recognize objects even when parts are missing.

Which augmentation is most similar in purpose to Random Erasing?

A Cutout

B Image flipping

C Color jittering

D Rotation

Cutout also removes parts of the image randomly, similar to Random Erasing.

Explain what Random Erasing is and why it helps in training computer vision models.

Think about how hiding parts of an image can help a model learn better.

Describe the key parameters that control how Random Erasing works.

Consider what controls how often and how big the erased parts are.