PreProcessing

Image Normalization

We'd like in this process for each feature to have a similar range so that our gradients don't go out of control (and that we only need one global learning rate multiplier).

SO what we do is we take average over all the images over all the pixels independently of each channel.

Question is:

• Why not over all the images independently for each pixel and each channel. This will basically give you an mean image to be subtracted from over all the images.

  • Because all pixels of the image share same weights and biases, hence it's important that we also take average over all pixels (in spatial dimension) as well.

• Is taking average over all the pixels in an image is a good idea? Doesn't it mess up inter contrastive features of the image for each channel?

  • May be not, because contrastive