Share input among all inference models within Ensemble pipeline

[adrtsang]

Hi, I have a question about how an input image volume to the pre-processing of the ensemble pipeline be accessed in other downstream models? For example, suppose the ensemble pipeline includes:
pre-processing -> model_1 inference -> model_2 pre-processing -> model_2 inference -> post-processing
what is the approach for model_2 inference to access the original input image volume to the pre-processing step?