From the course: AI Workshop: Hands-on with GANs with Deep Convolutional Networks

Unlock this course with a free trial

Join today to access over 23,300 courses taught by industry experts.

Viewing GAN training results

Viewing GAN training results

From the course: AI Workshop: Hands-on with GANs with Deep Convolutional Networks

Start my 1-month free trial Buy for my team

Viewing GAN training results

“

We set up the code for the training loop. All we have to do is run this code and let's see how our GAN performs. Now here we are in the very first epoch, epoch 0. This is where the results are completely random, both the generator and discriminator weights are completely untrained. Now let's iterate through a few epochs and then take a look at how these perform. The next output will be at around 200 batches of training. We are still on epoch 0, though you can't really see the epoch there. If you look at the generated images now, you can no longer call them complete noise. The generator has improved tremendously with just 200 iterations of training, but it's still pretty bad. The discriminator's total loss is 0.412. It's still low. The generator loss is 7.63, quite high. The real score of the discriminator is close to one, the fake score is close to zero, indicating the discriminator is doing a good job. The fake score after discriminator update is also close to zero, so the…

Contents