Choosing best technique for my project

I have created a tool that will fit an arbitrary 2d polygon geometry shape inside another arbitrary shape in 2d. It is very slow and I would love to use what I have learned from Andrew’s classes to make it fast using ml. I am looking for advice as to the best architecture to start investigating.

These images demonstrate the type of training data I can generate with the tool.
(Features X1)-Gray polygon model - The geometry to fit
(Features X2)-White N sided outline - The geometry to fit into
(Target Transform Y)-Black polygon model - The gray model transformed inside the white outline.


Any ideas? I could render X1 and X2 into images and use tensorflow to train those to Y (the target transform (translate,rotate,scale))?

The only thing I can thing of similar are bounding boxes in YOLO algorithm! You can check that out in Deep Learning Specialisation, or something similar to image segmentation like UNET, I guess!

Are these Delaunay triangulations? Why do you need NNs here if the solution is exact? Maybe it would be better to look at how to optimize your code like using multithreads or GPU.

Thank you. I will look into those.

Thank you for the response. They are not Delaunay. I am mostly doing this project for practice.