How to get an empty ConvLearner for single image prediction?

Here you go: