Inference results is inconsistent when using monai-trained model with ITK lib in C++

[7oud]

Is your feature request related to a problem? Please describe.
I pre-process a image with ITK in C++, then use MONAI-trained model to predict, but the result is inconsistent with python. I find the data layout is different, dimension order in ITK and pytorch/tensorrt is DHW, but MONAI is WHD, so before and after inference, two additional transpose is needed. That's inconvenient and low performance. When deploy a model in medical imaging, ITK may be indispensable. so is there any method to train a DHW data order model in MONAI?

Describe the solution you'd like
Or could give a data order option to choose when using MONAI?