When I train my model I train my Keras model on values 0.0-1.0 as floats, but anything except kCVPixelFormatType_OneComponent8 in CVPixelBufferCreate will create an error!
So my question is, should I start training my model with 0-255 as the input format instead even though I know I will get worse result? Or will CoreML convert unsigned chars to normalised floats behind my back?