Extension of model configurations for generative models

[RichardObi]

Hello 👋
Today @Warvito @nvahmadi and myself talked about the integration of generative models into MONAI Core. One aspect we discussed are pretrained generative models hosted in the Model Zoo. These models can have multiple inputs and outputs. For example, we could have a model with one image as input and a list of images as output.

This thread is to start a discussion to learn about ways of bringing pretrained generative models to the model zoo, and to possibly suggest configurations for them.

A second question is whether the current model zoo (and the bundle structure) could in the future also allow to add a framework of models instead of a single model to the model zoo. For example, we are thinking of bringing the medigan library to MONAI. medigan currently consists of 22 generative models and it seems it would be a nice integration if we could add the library "as a single model instance" to the model zoo. When users run the library via the model zoo, user could specify in the inference call which of the models in medigan is to be used to generate data. Could something like this be possible?