Using CSBDeep ImageJ plugin for custom DL network

Is it possible to use CSBDeep plugin to process an ImageJ hyperstack ​with a custom DL network previously trained in Tensorflow/Keras (Python)?

What are the constraints on the dimensions of the input (e.g. 3D multi-channel) and output (e.g. multiple 3D outputs)?

Is there a tutorial demonstrating how to proceed?

Hi @Sebastien, it’s possible in principle if you export the neural network from Python in a format that the CSBDeep Fiji plugin understands. There has been quite some effort around the standard, but I don’t know what the current status of support is within CSBDeep. Also, the same goes for multiple outputs. For both cases, @frauzufall probably knows the answer.