marcniethammer
commented
1 week ago Are you talking about inference or training? For inference, there are no parameters other than the number of instance optimization (IO) steps and you can pick the similarity measure most appropriate for your task for IO. That being said slightly better performance might be achievable by training with optimized parameters or adapting parameters for similarity measures. But we purposefully did not do this so far as our goal was to provide a generic registration approach that works out of the box without too much additional trouble.
Thank you very much for your work. Could you please give me your better parameter configuration for different data sets? I am using your model for CT registration, so I would like to know the parameters of your DirLab