3
$\begingroup$

I am trying to develop a model based on one-class classification approach. For example, the model should to identify if a given picture contains a cat or not. Keep in mind that my training dataset only contains pictures of cats and nothing else. Sort of like a anomaly detection problem

I have seen some suggestions on using autoencoders for unary classification but I couldn't find a concrete example on the net? Can someone point to an implementation of this approach using Keras. Is there any other way to about solving this problem?

$\endgroup$
1
  • 1
    $\begingroup$ you can find a very good reference here for Keras and if needed you can check this tutorial $\endgroup$ – Anurag Upadhyaya Oct 27 '17 at 4:10
1
$\begingroup$

If you have images of cats only, you could create boundary boxes (BB) of your images. Some BB will have cats an others won't. You will label those BB with cats inside as 1 as the others as 0.

This way you can set up a dataset with a binary class. It will be much easier if you already have boundary boxes for the cats in each image since this way it will be possible to generate the dataset automatically (generate random BB and label them as 1 if the intersection with the cat is big enough and 0 otherwise).

There is some software that helps to create boundary boxes.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.