search

diegoalejogm / gans

Generative Adversarial Networks implemented in PyTorch and Tensorflow
MIT License
832 stars 353 forks source link

Some questions #1

Closed mattbasta closed 3 years ago

mattbasta commented 6 years ago

Hi there. I'm very interested in creating GANs but have been unable to find much information, so I was hoping you could help answer somethings. Just a couple quick questions.

  1. For your TF implementation, you didn't use tf.contrib.gan. Can you briefly talk about why? Is it too restrictive?
  2. If, instead of training on and generating images, you wanted to create a GAN that trained on and generated a few dozen floating point numbers, how would that be accomplished? Is it simply a matter of changing the shape of the input to the discriminator, changing IMAGE_SIZE to be the number of floats, and supplying appropriate training data (from, say, a CSV)?

Thanks very much, your time is appreciated!

diegoalejogm commented 6 years ago

Sorry for the delay here,

@mattbasta :

  1. I wanted to create GANs from scratch, in order to be very clear on how they worked. I understand that tf.contrib.gan_ is an API that does implement GANs but hides all the technical details.

  2. Yes, that's exactly what you would do. Take into account that a 1D array must be the output, such as in the Vanilla GAN notebooks. If you use DC GAN, a 2D vector output will result, and it may not make sense for your purpose.

Hope I've responded to all your doubts successfully.