How to properly set up a neural network to highlight characteristic points?

Question

Pattern recognition

Good day, there is a task to find key points on faces
At the moment there is a certain number of input images (faces) and the same number of output face images with key points marked on them (points were marked with a marker in the photo editor)
It is planned to use the keras library for building a convolutional neural network due to good performance and detailed documentation
The question is:
What is the best way to feed images to CNN (black and white, grayscale, reduced, original size?) and what to give CNN as training output: images with marked points or an array with the coordinates of each point (there are about 70 of them)?

Answer 1

What is the best way to submit images to CNN (black and white, grayscale, scaled down, original size?)

In the form in which you later, at the stage of using your network, you will feed it images to find the right points.

and what to give the CNN as training output: images with marked points or an array with the coordinates of each point (there are about 70 of them)

Option 2 - "array with coordinates of points". Of course, if your goal is to "find key points on faces."