I'm trying to make a model for outputing an image based on speed (a float input). This can for example range from 1.0 to 6.0. so basically it is an cGAN, with a latent noise and a float as combined input.
I wonder what's the best way to encode my speed input? Currently I'm using embedding layers to encode the float to categorical data, but converting a sequence of float number to categorical/ordinal data seems dumb to me...but is this the only way to do it? I also want to extropolate data for prediction...(predict image from speed 0.5, or 6.2 for example) If I do categorical, then the model may not be able to predict something beyond my training data.
Many thanks!!