1
$\begingroup$

i am having an idea for a single-class classifier. I don't know if this is a logical "short circuit", though. The idea is the following:

Instead of a noise vector, i use a "noise-image" as input for training the GAN. Assuming i have a trained network, the classifier would work as follows. For example, the GAN should generate images of apples.

Class 1: An image of an apple is used as the input for the discriminator. Discriminator says: It's an original image The same image of an apple is used as the input for the generator. --> the generated image is an apple again. --> the discriminator says again: It's an original image --> The difference of both discriminator-outputs is low.

Some other class (here: dog, it represents the "not-apple" class) An image of a dog is used as the input for the discriminator. Discriminator says it is a "fake" image. An image of a dog is used as the input for the generator --> the generated image is an apple. --> the discriminator says it is an original image. --> The difference of both discriminator-outputs is high. --> This means the input image is NOT an apple.

Is this approach legit? Or does it have some logical flaws? I would like to hear what you think about it!

$\endgroup$

0

Your Answer

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

Browse other questions tagged or ask your own question.