I have a case where I want to feed a network with polylines of data. The problem is that the input can be any number of polylines and the polylines can consist of any number of points. If we instead convert the polylines to vectors the requirements might be easier to formulate:
- Network with a variable number of input vectors.
- All vectors have a variable length (independent).
- The input is NOT sequential.
- For one set of input only one output is requested (i.e not very similar to NLP/translation)
Any suggestions? I know that I can provide the network with a zero-padded 2D vector with the maximum number of "polylines" and the maximum number of "points" in each polyline as size but I would prefer not to since they can vary a lot.
I have previously converted the polylines to images of fixed size with success but I am looking for an alternative way to do it without the loss of precision that comes from converting from polylines to an image.