I've trained cntk model and now I want to evaluate it in my c++ code using CNTKLibrary. Configuration for training had this line:
featNorm = features - 128
p1 = MyLayer (featNorm, 32, 0.1557/256)
Does this mean that I have to deduct 128 from pixel values when I create NDArrayView to pass as input data for evaluation, or it is written somewhere in model and CNTKLibrary will normalize input data itself?
Because the normalization was done as part of the graph, you don't need to subtract 128 in your C++ code.