How to organize your base for searching by photos?

Question

Neural networks

Hey!
The topic of face recognition & detection has become interesting . Who is in the subject, tell me how you can organize your database for search?
For example, I have links with photos, and a unique identifier is stored in the link. What to do next, take 100 links and save one photo to disk and then scan it head-on? Or is it possible to build a hierarchy by features, first scanned a photo for search, determined some starting points, and then, how did I start searching for only images with these properties by index?
Who has experience in such matters, maybe also throw in some good literature :)
And who generally uses what: openCV or libraries in python, nodejs, go?

Answer 1

There are two tasks here:
1) encoding a picture with a vector - either singular points (SIFT and others), or a neuron (Triplet loss);
2) search - depends on the size of the database, a fullscan for a million takes about 50 ms, if this is a lot, then hnsw, annoy, faiss and others come to the rescue.