EasyDeepLearn

What is InstanceNorm and where does it shine?

medium

Answer

  • Normalize each channel independently within one example.
  • Removes per-image style / contrast information — hence its use in style transfer and image-to-image translation (CycleGAN, StyleGAN).
  • The scale/shift then allow the network to inject any desired 'style'.
  • Not usually used for classification because it destroys instance-specific information that helps discrimination.
Check yourself — multiple choice
  • InstanceNorm is used for image classification
  • Normalize each channel per example — removes style info, standard in style transfer / GANs
  • InstanceNorm depends on batch size
  • Same as BatchNorm

InstanceNorm strips style — that's why StyleGAN builds on it.

#normalization#generative

Practise Deep Learning

214 interview questions in this topic.

Related questions