Answer the question
In order to leave comments, you need to log in
Is there a ready-made neural network library or application that selects photos with an image of something given?
A friend had an interesting question, but since he does not have an account on Habré, I will work as a proxy. In order not to burden everyone with the subject area, I will formulate an equivalent problem. There is a database of photos, there are many of them. Of these, you need to sort out those where, say, a bicycle is photographed (in front, in profile, with or without a rider, or only, say, the front wheel with a steering wheel). There are a lot of samples for training a neural network, both with and without a bicycle. Is there a library / API / script / software that can be fed jpegs, and it will give out that in this photo there is a bicycle with a probability of 95%?
Answer the question
In order to leave comments, you need to log in
In order to stupidly feed a photo to a neural network and for it to find a bicycle on it, you need to have a neural network with a number of neurons comparable to the number of neurons in the human brain. And a quantitatively similar set of images for her training.
Neural networks are now used in practice for the tasks of recognizing small pre-prepared images. A bicycle in an arbitrary photo cannot be recognized in this way.
We need to dig in the direction of algorithms like face recognition, that is, an algorithmic search for elements and their characteristic location relative to each other.
As for the finished library, I don’t know, but you can look for a theory on the topic “bag of words”. Here, for example, there is a review in Russian, giving a general idea synthesis.ipi.ac.ru/sigmod/seminar/Vassilieva.ppt .
If it's not difficult, send your friend's contacts in a private message.
I have been interested in neural networks for a long time, maybe I can suggest something.
(My language is C)
Didn't find what you were looking for?
Ask your questionAsk a Question
731 491 924 answers to any question