Essay

Explain how synthetic blur can reduce a training–validation mismatch

Question: A company is building a road-sign classifier. Its training images come from sharp online photos, but its validation set comes from dashboard cameras in moving cars, so many validation images are blurred. Explain why this difference hurts performance and how adding synthetic motion blur to the training images can help.

Sample answer: The sharp online photos and the blurred dashboard-camera images come from different distributions. Because the validation set contains motion blur caused by camera movement in a moving vehicle, the model may not learn patterns that hold up on blurred inputs if it only sees crisp training images. One practical fix is to apply simulated motion blur to some of the sharp training photos. That makes the training data more like the validation data, reduces the mismatch between the two sets, and usually improves validation performance.

Key points:

  • The validation images are blurred because the camera is moving in a car.
  • The training images are sharp online photos.
  • Synthetic motion blur is added to the training images to better match the validation distribution.

Rubric: Answers must explain that the validation set is blurred due to camera motion, the training set contains sharp online images, and synthetic motion blur is applied to the training images to make them resemble the validation set and reduce the distribution gap.

0

1

Updated 2026-08-12

Contributors are:

Who are from:

Tags

Machine Learning

Deep Learning

Supervised Learning

Dive into Deep Learning @ D2L

Data Science

Machine Learning Strategy

Machine Learning Yearning @ DeepLearning.AI