I have been using Windows.AI.MachineLearning.Preview, after switching to Windows.AI.MachineLearning and importing my .onnx model. I realised that I can now evaluate images that have the specific size of 227 by 227.
This doesn't work for my application as I don't know the size of the image and I can't crop or resize the image as it might change what it represents.
With Windows.AI.MachineLearning for build 17763 you have a couple of options:
Does that work for you ?