Tiramisu and batchnorm in new keras api

I am trying to run the script for the tiramisu model in:
https://github.com/jph00/part2/blob/master/tiramisu-keras.ipynb
but I am running with errors due to changes in functions with the latest keras (e.g. Deconvolution2D is now Conv2DTranspose, etc).
Most of them can be easily solved except for BatchNormalization(mode=2, axis=-1). the mode option is no longer available. Does anyone know how to translate this mode option to the new function?

2 Likes

You can fix to Keras by removing mode=2. Most of the function calls are little syntax things. The only real different change is the generator I think, but it will still be backwards compatible for a while.