Since the addition of the compatibility with Keras 2, the call to merge() in concatenate_layers() no longer includes the mode parameter when used with Keras 1. Thus, merge() is now called with the default mode='sum' default value instead of with 'concat', which I understand that is not the expected behaviour (and crashes if the number of channels is different for A and B).
Hi,
Since the addition of the compatibility with Keras 2, the call to merge() in concatenate_layers() no longer includes the
mode
parameter when used with Keras 1. Thus, merge() is now called with the defaultmode='sum'
default value instead of with 'concat', which I understand that is not the expected behaviour (and crashes if the number of channels is different for A and B).Many thanks for sharing this work!
Best,
Aitor