How to do training with segmentation?

Question

Neural networks

There is no experience with neurons yet, and it is not clear where to dig.
What can be used to make a neuron and train it by feeding X cards and for each picture indicate the bool value whether something is present on it. For example, let's take pictures of doors, and bool is the value of whether there is glazing at the door. And then upload a new picture and determine:
- whether there is a door in the picture
- determine whether the door has glazing
- after that, find from our database on which we studied, the most similar doors from this segment (with or without glazing)

I would like to hear advice on choosing a tool and preferably links to some articles with an analysis of similar tasks. Thank you!

Answer 1

If you have trained the network to answer Да/Нет, you will never get similar doors. To do this, the network must convert the image into a vector, and then you can find similar vectors. Absolutely the same as doing a search by photo.