What structure should be used to build points on the face?

Question

Pattern recognition

Good day, there is a selection of 300 photographs (taking into account augmentation, it was no longer possible to squeeze out) images of faces and 300 of the same faces, only with marked reference points
The task is to teach the machine to find these very points on any face
There was no experience in this, but obviously it's a matter of convolution, but how to competently build a process and train a machine?