CSBDeep StructN2V - prediction doesn't work

I can train and predict using the Fiji N2V plugin on my images, but most of the noise remain intact because noise is structural.
I tried the provided Jupyter Notebook for training (after all required Python installations) and can train the StructN2V network on my images and save the model. But the provided Jupyter Notebook for predicting fails with the following error: “UnknownError: 2 root error(s) found.
(0) Unknown: Failed to get convolution algorithm. This is probably because cuDNN failed to initialize,…”
When I tried to predict using Fiji and the saved model, I got another error: “java.lang.IllegalArgumentException: 2 root error(s) found.
(0) Invalid argument: You must feed a value for placeholder tensor ‘input_1’ with dtype float and shape [?,?,?,1] …”

HI Vladimir, welcome to image.sc

It looks like your CUDA/cuDNN installation is faulty. Can you double check that you installed everything correctly there.
The Fiji error sounds like there is no open input imge to do the prediction on.

@colemanbroad @akrull @uschmidt83 @mweigert should be able to provide more detailed help.