In Keras backends document, image_dim_ordering controls how to interpret the information of image shape and input channel.
In current source code, keras.backend.backend() is used as the condition instead of keras.backend.image_dim_ordering(). TensorFlow backend is possible to use th as its dimension ordering convention, and vice versa.
jakebian
commented
7 years ago
In Keras backends document,
image_dim_orderingcontrols how to interpret the information of image shape and input channel.In current source code,
keras.backend.backend()is used as the condition instead ofkeras.backend.image_dim_ordering(). TensorFlow backend is possible to usethas its dimension ordering convention, and vice versa.