For BatchNormalization function which is used in the Densenet implementation, the axis parameter should be the channel axis. For different data formats, we should assign different values to the parameter. Concretely, the channel is -1 (last axis) for Tensorflow while 1 (second axis) for Theano.
For BatchNormalization function which is used in the Densenet implementation, the axis parameter should be the channel axis. For different data formats, we should assign different values to the parameter. Concretely, the channel is -1 (last axis) for Tensorflow while 1 (second axis) for Theano.