search

ylf-li / ContourGAN

Tensorflow for ContourGAN: Image Contour Detection with Generative Adversarial Encoder-Decoder Networks
73 stars 16 forks source link

No use of Discriminator? #8

Open aditipanda opened 5 years ago

aditipanda commented 5 years ago

In the file demoTest.py that you released recently, there is no call to discriminator. Only the generator outputs are compared with ground-truth and the corresponding loss is minimized.

But this is a GAN network. Both the generator and discriminator need to be used. Kindly help me understand.

doantientai commented 5 years ago

The Generator in this paper works as an edge detector. The Discriminator helps the Generator to learn better in the train phase. However, in the test phase, we test the ability to detect edges of the Generator, so Discriminator won't be used at all. This is the same for most GANs.