Apache Spark Deep Learning Cookbook

There's more...

Consider the following:

• Once the images are done preprocessing they still need to be normalized and converted into vectors (in this case tensors) before being fed into the network.
• Normalization, in the simplest case, means adjusting values measured on different scales to a notionally common scale, often prior to averaging. It is always a good idea to normalize data in order to prevent gradients from exploding or vanishing as seen in the vanishing and exploding gradient problems during gradient descent. Normalization also ensures there is no data redundancy.
• Normalization of the data is done by dividing each pixel in each image by 255 since the pixel values range between 0 and 255. This will result in the output shown in ...