(Tensorflow) Parameter prediction from image

Question

First I will give some context of my situation:

I have a laser that after propagation is observed by a camera. Now the initial parameters of the laser will determine the pattern observed in the image taken by the camera. I want to train a NN that takes as input the image of the laser after propagation and is capable to output the 6 initial parameters that (in my case) cause the observed shape. As training data, I have already a simulation that generates the pattern that should be observed for a set of initial parameters and also some experimental data.

My problem is while I know some Tensorflow/keras already (like image classification), i am not sure how to go from an input image to an output vector with non-discrete values. Could anybody give me some indications or some references for similar cases ?

Sorry if it is a really simple question.

Thank you in advance.

I was thinking of instead of outputting a vector with values, to output an image with the values of each px representing each initial parameter value prediction. However, I think is probably much more complicated than a just doing the vector.

I searched on the internet and google scholar but couldn't find anything.