Preprocessing in examples

[georgedeath]

If using the, for example, VGG19 backend, the corresponding keras guide recommends applying the preprocessing from here:

Note: each Keras Application expects a specific kind of input preprocessing. For VGG19, call tf.keras.applications.vgg19.preprocess_input on your inputs before passing them to the model. vgg19.preprocess_input will convert the input images from RGB to BGR, then will zero-center each color channel with respect to the ImageNet dataset, without scaling.

The examples show only rescaling [0, 255] to [0, 1], when, in actual fact, this is nothing like the images the backend networks were trained on. When using the correct scaling (i.e. via preprocess_input) training speed is fastly improved, at least in my experience.

[yingkaisha]

@georgedeath Thank you for pointing it out. I will experiment with this preprocessing option and update the example accordingly.